MoleculeMind

副标题：全球临床管线、核心公司、七大技术路线与资本格局全析Subtitle: Global Clinical Pipelines, Core Companies, Seven Technical Routes, and Capital Landscape Analysis 发布日期： 2026年3月1日Release Date: March 1, 2026 报告类型：行业深度研究报告（行业/板块研究）Report Type: In-depth Industry Research Report (Industry/Sector Research) 目标读者：专业投资者、行业分析师、机构研究人员Target Audience: Professional Investors, Industry Analysts, Institutional Researchers 数据截止日期： 2026年2月28日Data Cut-off Date: February 28, 2026 覆盖公司： Insilico Medicine (3696.HK)、Recursion Pharmaceuticals (RXRX)、Isomorphic Labs、Relay Therapeutics (RLAY)、Xaira Therapeutics、Schrödinger (SDGR)、BenevolentAI (BAI)、晶泰科技 (2228.HK) 及其他Covered Companies: Insilico Medicine (3696.HK), Recursion Pharmaceuticals (RXRX), Isomorphic Labs, Relay Therapeutics (RLAY), Xaira Therapeutics, Schrödinger (SDGR), BenevolentAI (BAI), XtalPi (2228.HK), and others. 免责声明----------Disclaimer 本报告仅供专业投资者和行业研究人员参考，不构成任何投资建议或买卖任何证券的邀约或要约。报告中所有数据、预测及分析均来源于公开信息，包括公司招股书、监管文件、学术论文及第三方研究机构报告，准确性和完整性不做保证。投资者在做出任何投资决策前，应自行进行独立研究并咨询持牌专业投资顾问。报告中涉及的前瞻性陈述存在固有风险和不确定性，实际结果可能与预测存在重大差异。医药行业具有特殊风险，临床试验结果具有高度不确定性，监管批准不能保证。本报告编制者不持有本报告所涵盖公司的任何证券头寸，亦不存在影响报告客观性的利益冲突。过往业绩不代表未来结果。This report is for the reference of professional investors and industry researchers only and does not constitute any investment advice or an offer or solicitation to buy or sell any securities. All data, forecasts, and analyses in the report are derived from public information, including company prospectuses, regulatory filings, academic papers, and third-party research reports; accuracy and completeness are not guaranteed. Investors should conduct independent research and consult licensed professional investment advisors before making any investment decisions. Forward-looking statements in the report involve inherent risks and uncertainties; actual results may differ materially from forecasts. The pharmaceutical industry carries specific risks; clinical trial results are highly uncertain, and regulatory approvals cannot be guaranteed. The preparers of this report hold no securities positions in the covered companies and have no conflicts of interest that would affect the report’s objectivity. Past performance does not indicate future results. 执行摘要--------Executive Summary核心论点Core Argument 2026年是AI制药从概念验证走向临床证据的分水岭。自2020年全球首个AI设计药物进入人体试验以来，行业已历经六年积累——管线扩容、平台迭代、资本涌入——但始终面临一个核心质疑：AI设计的分子，真的能治病吗？2026 is the watershed moment for AI drug discovery, transitioning from proof-of-concept to clinical evidence. Since the first AI-designed drug entered human trials in 2020, the industry has experienced six years of accumulation—pipeline expansion, platform iteration, capital influx—yet it has always faced a core query: Can AI-designed molecules truly treat diseases? 2025年6月，一篇发表于《自然·医学》的论文正式给出了第一个经随机对照临床试验验证的答案。英矽智能（Insilico Medicine）的AI药物Rentosertib（ISM001-055）在特发性肺纤维化（IPF）的Phase 2a试验中显示出统计学显著的疗效信号：60mg QD组用力肺活量（FVC）改善+98.4 ml，而安慰剂组下降-62.3 ml。这是全球首个完成随机对照Phase 2a临床试验的AI原创设计小分子药物，标志着行业进入**“临床验证元年”**。In June 2025, a paper published in Nature Medicineofficially provided the first answer validated by randomized controlled clinical trials. Insilico Medicine’s AI drug Rentosertib (ISM001-055) showed statistically significant efficacy signals in a Phase 2a trial for idiopathic pulmonary fibrosis (IPF): the 60mg QD group’s forced vital capacity (FVC) improved by +98.4 ml, while the placebo group declined by -62.3 ml. As the world’s first AI-originated small molecule drug to complete a randomized controlled Phase 2a trial, it marks the industry’s entry into the “First Year of Clinical Validation”.五大核心发现Five Core Findings 1. 全球AI临床管线规模已达173个项目（2026年初），其中Phase I 94个、Phase II 56个、Phase III 15个。AI药物发现的Phase 1成功率（80-90%）显著高于传统药物（40-65%），初步验证了AI在候选药物质量提升上的价值。1. The global AI clinical pipeline has reached 173 projects (early 2026), including 94 in Phase I, 56 in Phase II, and 15 in Phase III. The Phase 1 success rate for AI drug discovery (80-90%) is significantly higher than that of traditional drugs (40-65%), preliminarily validating AI’s value in improving candidate drug quality. 2. Rentosertib是最重要的行业催化剂。 2025年6月，Nature Medicine发表的Phase 2a数据证明了TNIK是IPF的可成药靶点，且AI原创发现的分子具有真实临床疗效。该药物于18个月内完成靶点发现到临床前候选的全过程（传统需4.5年），是端到端AI制药能力的最强证明。2. Rentosertib is the most critical industry catalyst. Phase 2a data published in Nature Medicine in June 2025 proved that TNIK is a druggable target for IPF, and the AI-originated molecule has real clinical efficacy. Completing the entire process from target discovery to preclinical candidate within 18 months (compared to the traditional 4.5 years), this drug serves as the strongest proof of end-to-end AI drug discovery capabilities. 3. AlphaFold系列技术商业化加速。 Google DeepMind旗下Isomorphic Labs于2025年3月完成6亿美元首轮外部融资，并宣布首批AI设计候选药物即将进入人体试验（2025年7月）。2026年初，"AlphaFold 4"级别新模型见诸报道，蛋白质结构预测技术持续领先并深入药物设计全链条。3. The commercialization of AlphaFold series technologies is accelerating. Google DeepMind’s Isomorphic Labs completed a $600 million initial external financing round in March 2025 and announced that its first batch of AI-designed drug candidates would soon enter human trials (July 2025). By early 2026, reports of “AlphaFold 4” level new models emerged, showing protein structure prediction technology continuing to lead and deeply integrate into the entire drug design chain. 4. 资本格局两极分化明显。2025年全球AI制药风险投资达57亿美元（同比+78%），但向头部集中：Xaira（超10亿美元A轮）、Isomorphic（6亿美元）、英矽智能（港股IPO募集22.77亿港元）获得史级规模融资；而BenevolentAI因临床失败市值较IPO高点下跌超90%，成为行业警示。4. The capital landscape is clearly polarizing. Global venture capital in AI drug discovery reached 1 billion** Series A), Isomorphic ($600 million), and Insilico Medicine (HK IPO raising HKD 2.277 billion) secured historic level financing; meanwhile, BenevolentAI’s market cap plummeted over 90% from its IPO peak due to clinical failures, serving as an industry warning. 5. 监管环境趋于明朗。FDA已接受500+份含AI成分的药物申请（接受率>95%），2026年Q2预计发布最终AI监管指南；NMPA在中国提供加速审评路径，默克中国区总裁预测中国或成为全球首个批准AI设计药物的市场。5. The regulatory environment is becoming clearer. The FDA has accepted 500+ drug applications containing AI components (acceptance rate >95%), with final AI regulatory guidelines expected in Q2 2026. The NMPA offers accelerated review pathways in China, and Merck’s China President predicts China may become the first market globally to approve an AI-designed drug.2026年关键催化剂Key Catalysts in 2026 催化剂 / Catalysts 预期时间 / Expected Time 影响评估 / Impact Assessment Rentosertib Phase 2b/3启动（全球多中心，IPF） Rentosertib Phase   2b/3 Initiation (Global Multicenter, IPF) 2026年H1-H2 2026 H1-H2 行业级 Industry-level RLY-2608 Phase III启动（乳腺癌PI3Kα） RLY-2608 Phase   III Initiation (Breast Cancer PI3Kα) 2026年H2 2026 H2 行业级 Industry-level REC-4881 FDA注册路径磋商（FAP） REC-4881 FDA Registry Path Consultations (FAP) 2026年H1 2026 H1 公司级 Company-level Isomorphic Labs首批临床数据公布 Isomorphic   Labs Initial Clinical Data Release 2026年 2026 行业级 Industry-level FDA最终AI监管指南发布 FDA Final AI Regulatory Guidelines Release 2026年Q2 2026 Q2 政策级 Policy-level FDA首个明确标注AI设计药物审评启动 FDA Review Initiation for First Labelled AI-Designed Drug 2026-2027年 2026-2027 里程碑级   Milestone-level主要风险提示Major Risk Warnings · 临床成功率不确定性：AI加速候选发现，但不改变后期临床失败率（整体~10%）· Uncertainty in clinical success rates: AI accelerates candidate discovery but does not alter late-stage clinical failure rates (overall ~10%).· 估值泡沫风险：多数公司距商业化收入仍5年以上· Valuation bubble risk: Most companies are still over 5 years away from commercial revenue.· 竞争加剧：大型药企内部AI能力快速提升，中间平台型公司面临被整合压力· Intensified competition: Internal AI capabilities of big pharma are rapidly improving, putting pressure on mid-tier platform companies to integrate.· 数据依赖性：AI模型质量高度依赖训练数据质量与数量· Data dependency: AI model quality is highly dependent on the quality and quantity of training data. 第一章：行业概况与市场规模-----------------Chapter 1: Industry Overview and Market Size1.1 市场规模：多机构预测汇总1.1 Market Size: Summary of Multi-agency Forecasts AI辅助药物发现市场的规模因各机构定义范围不同而存在较大差异。综合主要市场研究机构数据：The scale of the AI-assisted drug discovery market varies greatly due to differing definitions by various agencies. Synthesizing data from major market research firms: 机构 / Agency 2025年规模 / 2025 Size 2030-2035年规模 / 2030-2035    Size CAGR Precedence Research 69.3亿美元 $6.93 Billion 178.1亿美元（2035） $17.81 Billion   (2035) 9.9% Mordor Intelligence 25.8亿美元 $2.58 Billion 102.9亿美元（2031） $10.29 Billion   (2031) 25.8% Roots Analysis 29.0亿美元 $2.90 Billion 134.0亿美元（2035） $13.40 Billion   (2035) 11.3% Straits Research 24.0亿美元 $2.40 Billion 253.5亿美元（2034） $25.35 Billion   (2034) 29.9% The Business Research Co. 23.3亿美元 $2.33 Billion 74.2亿美元（2030） $7.42 Billion   (2030) — 共识区间 / Consensus   Range 25-70亿美元 $2.5-7.0   Billion 100-250亿美元 $10-25 Billion 10-30% 差异主要来源于定义范围：狭义（仅软件授权收入）约23-30亿美元；广义（含AI合作里程碑、AI赋能CRO等）可达60-70亿美元。无论使用哪个口径，行业增长趋势均高度一致：2025-2035年将实现至少3-5倍增长。Variations primarily stem from definitional scopes: narrowly defined (software licensing revenue only) is about 6.0-7.0 billion. Regardless of the metric used, the industry growth trend is highly consistent: it will achieve at least 3-5 times growth from 2025 to 2035. 亚太地区增速最快，CAGR达21.1%，以中国为核心驱动力（政策支持、人才红利、临床资源丰富）。The Asia-Pacific region is the fastest-growing, with a CAGR of 21.1%, driven essentially by China (policy support, talent dividend, abundant clinical resources). 图1：多机构市场规模预测对比Figure 1: Comparison of Market Size Forecasts by Multiple Agencies1.2 AI与传统药物发现的对比1.2 Comparison between AI and Traditional Drug Discovery 传统药物发现的痛点已是行业共识：The pain points of traditional drug discovery are an industry consensus: 指标 / Metric 传统方法 / Traditional    Method AI辅助/AI原创 / AI-Assisted/AI-Originated 改善幅度 / Improvement    Level 靶点→临床前候选周期 Target →   Preclinical Candidate Cycle 4-5年 4-5 years 12-18个月（Insilico实测） 12-18 months   (Insilico actual) 缩短60-70%Reduced by 60-70% 化合物筛选数量 Number of   Compounds Screened 2,500+（Recursion估算） 2,500+   (Recursion est.) ~330（Recursion实测） ~330 (Recursion   actual) 减少87%Reduced by 87% 单个候选发现成本 Cost per   Candidate Discovery 数亿美元 Hundreds of   millions USD 大幅降低（具体数据保密）   Significantly reduced (data confidential) 显著降低 Significantly   reduced Phase I成功率 Phase I Success   Rate 40-65% 80-90%（AI项目实测） 80-90% (AI   projects actual) 提升约40ppIncreased by ~40pp Phase II成功率 Phase II   Success Rate 30-45% 65-75%（AI项目实测） 65-75% (AI   projects actual) 提升约30ppIncreased by ~30pp 整体新药从发现到批准 Overall   New Drug Discovery to Approval 10-15年 10-15 years 预计节省3-5年 Expected savings   of 3-5 years 缩短25-35% Reduced by   25-35% 注意事项：以上AI的成功率数据来源于Axis Intelligence 2026年对173个AI临床项目的分析，整体样本量较小，且存在选择偏倚（AI公司更倾向于公布积极结果）。更多临床数据积累后，这些数据将更具说服力。Note: The above AI success rate data comes from Axis Intelligence’s 2026 analysis of 173 AI clinical projects. The overall sample size is relatively small, and there is selection bias (AI companies are more likely to publish positive results). Once more clinical data is accumulated, these statistics will be more persuasive. 图2：AI药物发现与传统方法时间线对比Figure 2: Timeline Comparison of AI Drug Discovery vs. Traditional Methods1.3 行业驱动因素1.3 Industry Drivers 技术驱动：Technology Drivers:· AlphaFold2/3革命性提升蛋白质结构预测精度（2020年、2024年）· AlphaFold2/3 revolutionized the accuracy of protein structure prediction (2020, 2024)· 大型生成AI模型（Transformer架构）从NLP迁移到分子设计· Large Generative AI models (Transformer architecture) migrating from NLP to molecular design· GPU算力成本下降使大规模分子模拟成为可能· Decreasing GPU computing costs enabling large-scale molecular simulations· 公开数据库（ChEMBL、PDB、UniProt）数据量持续扩大· Continuous expansion of data in public databases (ChEMBL, PDB, UniProt) 需求驱动：Demand Drivers:· 大型药企面临重磅炸弹药物专利悬崖，迫切需要补充管线· Big pharma facing patent cliffs for blockbuster drugs, urgently needing to replenish pipelines· 传统药物发现效率低下（Eroom’s Law：研发效率每9年降低一半）· Low efficiency in traditional drug discovery (Eroom’s Law: R&D efficiency halves every 9 years)· 罕见病、难成药靶点对AI方法有特殊需求· Rare diseases and “undruggable” targets present special needs for AI methodologies 政策驱动：Policy Drivers:· FDA积极接纳AI申请（>95%接受率），并建立专门工作组· FDA actively accepts AI applications (>95% acceptance rate) and establishes dedicated task forces· 各国政府加大对AI医疗的资助力度（美国NIH、欧盟Horizon、中国各省市）· Governments worldwide increasingly funding AI healthcare (US NIH, EU Horizon, Chinese provinces/cities) 第二章：2026年关键里程碑与行业背景-----------------------Chapter 2: Key Milestones in 2026 and Industry Background2.1 行业发展时间线（2020-2026年）2.1 Industry Development Timeline (2020-2026) 年份 / Year 事件 / Event 意义 / Significance 2020.01 DSP-1181（Exscientia）进入日本Phase 1（OCD） DSP-1181   (Exscientia) enters Phase 1 in Japan (OCD) 全球首个AI设计药物进入人体试验 World’s   first AI-designed drug enters human trials 2020.09 EXS21546（Exscientia）进入英国Phase 1（肿瘤免疫） EXS21546   (Exscientia) enters Phase 1 in UK (Immuno-oncology) 全球首个AI肿瘤免疫药进入临床 World’s   first AI IO drug enters clinics 2020.11 AlphaFold2发布，CASP14中达到实验级精度 AlphaFold2   published, achieves experimental-level precision in CASP14 蛋白质折叠问题被攻克，药物靶点发现革命   Protein folding problem solved, revolution in drug target discovery 2021 Exscientia、Recursion相继纳斯达克IPO   Exscientia, Recursion sequentially IPO on Nasdaq AI制药公司进入资本市场 AI drug   companies enter the capital market 2021 Isomorphic Labs从DeepMind分拆成立 Isomorphic   Labs formed as spin-off from DeepMind AlphaFold商业化正式启动   Commercialization of AlphaFold officially begins 2022 BenevolentAI阿姆斯特丹上市（估值10亿英镑） BenevolentAI   lists in Amsterdam (Valuation £1 Billion) 欧洲AI制药公司进入公开市场 European   AI drug companies enter public markets 2023 Insilico Medicine Rentosertib完成Phase 1 Insilico   Medicine Rentosertib completes Phase 1 全球首个AI原创IPF药物完成Phase 1 World’s   first AI-originated IPF drug completes Phase 1 2024.05 AlphaFold3发布（蛋白质-小分子相互作用预测）   AlphaFold3 released (Predicts protein-small molecule interactions) 结构生物学新里程碑 New   milestone in structural biology 2024.10 David Baker、Demis Hassabis、John Jumper共获诺贝尔化学奖 David   Baker, Demis Hassabis, John Jumper share Nobel Prize in Chemistry AI蛋白质科学获最高学术认可   Highest academic recognition for AI protein sciences 2024.11 Recursion完成收购Exscientia（约6.88亿美元） Recursion   acquires Exscientia (approx. $688 million) AI制药最大并购，行业整合加速   Largest M&A in AI drug discovery, accelerating industry consolidation 2024 Xaira Therapeutics成立，获超10亿美元A轮 Xaira   Therapeutics founded, receives >$1 Billion Series A AI制药史上最大A轮融资 Largest Series   A funding in AI drug discovery history 2025.03 Isomorphic Labs完成6亿美元首轮外部融资   Isomorphic Labs completes $600 million initial external financing AlphaFold商业化获顶级资本背书   AlphaFold commercialization endorsed by top-tier capital 2025.06 Rentosertib Phase 2a随机对照数据发表Nature   MedicineRentosertib Phase 2a RCT data   published in Nature Medicine 行业历史性里程碑：首个完成RCT的AI设计药物Historic   industry milestone: First AI-designed drug to complete RCT 2025.07 Isomorphic Labs宣布首批AI候选药物启动人体试验   Isomorphic Labs announces initial AI candidates starting human trials 又一AI原生公司进入临床 Another   AI-native company enters the clinic 2025.12 英矽智能港交所主板上市（3696.HK），募集22.77亿港元 Insilico   Medicine lists on HKEX Main Board (3696.HK), raising HKD 2.277 Billion “AI制药第一股”，2025年港股最大生技IPO “First   AI Drug Stock,” largest biotech IPO on HKEX in 2025 2026.01 默克中国区总裁预测中国或成首个批准AI设计药物的市场 Merck China   President predicts China might be the first to approve AI-designed drugs 监管信号重要 Important   regulatory signal 2026.02 Recursion获赛诺菲第5次里程碑（400万美元），REC-4881 FAP数据积极 Recursion   receives 5th milestone from Sanofi ($4 million), REC-4881 FAP data positive Recursion平台价值进一步验证 Recursion   platform value further validated 2026.Q2（预期） FDA最终AI监管指南发布 FDA Final AI   Regulatory Guidelines release (Expected) 监管框架成熟关键节点 Crucial   node for regulatory framework maturation 图3：AI制药行业里程碑时间线Figure 3: Timeline of AI Drug Discovery Industry Milestones2.2 监管环境全景2.2 Overview of the Regulatory Environment FDA（美国）FDA (United States) FDA已在AI制药监管上走在全球前列：The FDA is at the global forefront regarding the regulation of AI in pharmaceuticals:· 接受率：截至2025年，500+份含AI成分的药物申请中，>95%已被接受· Acceptance Rate: As of 2025, out of 500+ drug applications containing AI components, >95% have been accepted.· AI提交量年增40%（2021-2023年数据）· AI submissions growing by 40% annually (2021-2023 data).· 2025年1月：发布草案"AI支持监管决策的考量"，收到500+公众意见· January 2025: Issued the draft guidance “Considerations for Using AI to Support Regulatory Decision-Making,” receiving 500+ public comments.· 2026年Q2：预计发布最终指南· Q2 2026: Final guidelines expected to be published.· NAM路线图（2025年4月）：支持非动物方法（与AI预测互补）· NAM Roadmap (April 2025): Supports non-animal methods (complementary to AI predictions). FDA的核心监管关切：FDA’s core regulatory concerns:1. AI模型的可验证性与可重现性 1.Verifiability and reproducibility of AI models.2. 训练数据集的代表性与偏差 2.Representativeness and bias in training datasets.3. AI在关键决策中的透明度 3.Transparency of AI in critical decision-making.4. 提交材料中AI应用的完整记录 4.Comprehensive tracking of AI applications in submissions. EMA（欧盟）EMA (European Union)· AI资格认定框架开发中，预计2026年Q2发布· AI qualification framework under development, expected for release in Q2 2026.· PRIME加速审评计划已可适用于AI辅助发现的罕见病药物· The PRIME expedited review scheme is already applicable to AI-assisted drug discovery for rare diseases. NMPA（中国）NMPA (China)· 加速审评路径已对AI辅助发现药物开放· Accelerated review pathways are open to AI-assisted discovered drugs.· 英矽智能Rentosertib有望在中国同步推进Phase 2b/3全球试验· Insilico Medicine’s Rentosertib is poised to advance its Phase 2b/3 global trials concurrently in China.· 中国被多方预测为全球最可能率先批准AI设计药物的市场· China is widely predicted to be the most likely market globally to approve an AI-designed drug first. ICH（国际）ICH (International)· 2025年启动AI专题指南开发，预计2027年完成· Development of AI-specific guidelines initiated in 2025, expected to be completed in 2027. 第三章：全球临床管线全析----------------Chapter 3: Global Clinical Pipeline Analysis3.1 规模概览3.1 Scale Overview 根据Axis Intelligence 2026年完整分析，全球AI相关临床药物项目：According to a complete 2026 analysis by Axis Intelligence, global AI-related clinical drug projects are as follows: 阶段 / Phase 项目数量 / Project Count AI药物Phase成功率 / AI Drug Phase Success Rate 传统药物Phase成功率 / Traditional Drug Phase Success Rate Phase I 94个 / 94 80-90% 40-65% Phase II 56个 / 56 65-75% 30-45% Phase III 15个 / 15 待观察 / To be observed 50-65% 合计 / Total 173个 / 173 — — 图4：AI临床管线阶段分布Figure 4: Distribution of AI Clinical Pipeline Stages 适应症分布：Indication Distribution: 适应症领域 / Indication Area 占比 / Share Phase I成功率 / Phase I Success Rate Phase II成功率 / Phase II Success Rate 肿瘤学 / Oncology 45% 80% 64% 纤维化疾病 / Fibrotic Diseases 14% 88% 75% 神经退行性疾病 / Neurodegenerative Diseases 10% 72% 44% 传染病 / Infectious Diseases 10% 94% 82% 代谢性疾病 / Metabolic Diseases 8% — — 罕见病 / Rare Diseases 6% 91% — 其他 / Others 7% — — 图5：AI临床药物适应症分布Figure 5: Distribution of AI Clinical Drug Indications3.2 重点药物深度分析3.2 In-depth Analysis of Key Drugs 3.2.1 Rentosertib（ISM001-055）——行业最重要里程碑3.2.1 Rentosertib (ISM001-055) — The Industry’s Most Important Milestone 基本信息Basic Information 属性 / Attribute 详情 / Details 开发公司 / Developing Company Insilico   Medicine（英矽智能） 药物类型 / Drug Type 口服小分子，first-in-class / Oral small   molecule, first-in-class 靶点 / Target TNIK（Traf2- and   Nck-interacting kinase） 适应症 / Indication 特发性肺纤维化（IPF） / Idiopathic   pulmonary fibrosis (IPF) 当前阶段 / Current Phase Phase 2a已完成，规划Phase 2b/3 /   Phase 2a completed, Phase 2b/3 planned ClinicalTrials   ID NCT05938920（Phase 2a）；NCT05154240（Phase 1） AI发现过程的历史意义Historical Significance of the AI Discovery Process Rentosertib是首个由AI同时完成靶点发现和先导化合物设计的药物：Rentosertib is the first drug where AI simultaneously accomplished both target discovery and lead compound design:· Biology42平台：从多组学数据中识别TNIK为IPF的关键节点激酶（此前非已知靶点）· Biology42 Platform: Identified TNIK as a critical nodal kinase in IPF from multi-omics data (previously an unknown target).· Chemistry42平台：在18个月内生成并优化出候选化合物ISM001-055· Chemistry42 Platform: Generated and optimized the candidate compound ISM001-055 within 18 months.· 全程耗时：18个月（靶点→临床前候选），传统方法需4-5年· Total elapsed time: 18 months (target → preclinical candidate); traditional methods require 4-5 years. Phase 2a临床结果（2025年6月，Nature Medicine）Phase 2a Clinical Results (June 2025, Nature Medicine) 试验设计：多中心、双盲、随机、安慰剂对照；21个中心（中国）；12周随机治疗期。入组71例IPF患者。Trial Design: Multicenter, double-blind, randomized, placebo-controlled; 21 centers (China); 12-week randomized treatment period. Enrolled 71 IPF patients. 治疗组 / Treatment Group FVC均值变化（ml） / Mean Change in FVC (ml) 95%    CI 60mg QD（每日一次） / (Once   Daily) +98.4 10.9   ~ 185.9 30mg BID（每日两次） / (Twice   Daily) +19.7 -60.5 ~   99.9 30mg QD -27.0 -88.8 ~   34.8 安慰剂 / Placebo -62.3 -116.1 ~   75.6 数据说明（核查注释）：安慰剂组数据存在不确定性。研究参考文件注明，“-62.3 ml"可能来自二次分析报告，与Nature Medicine原文数据（可能为-20.3 ml）存在差异；本表中CI已更新为”-116.1 ~ 75.6"（与研究参考文件一致），该CI与-62.3 ml数学自洽（见原版报告中的-116.1 ~ -8.5）。建议在正式发布前对照Nature Medicine原文（DOI: 10.1038/s41591-025-03743-2）核实安慰剂组均值和置信区间。 60mg QD组数据（+98.4 ml，CI: 10.9~185.9）与所有来源完全一致，不受影响。Data Clarification (Fact-Check Note): Uncertainty exists in the placebo group’s data. Reference research notes that “-62.3 ml” might have originated from a secondary analysis report, varying from original Nature Medicine text data (possibly -20.3 ml). The CI in this table has been updated to “-116.1 ~ 75.6” (aligned with reference documents), perfectly mathematically compatible with -62.3 ml (vs. original report’s -116.1 ~ -8.5). It is recommended to verify the placebo mean and CI against the original Nature Medicine paper (DOI: 10.1038/s41591-025-03743-2) before formal release. 60mg QD group data (+98.4 ml, CI: 10.9~185.9) remains fully consistent across all sources untouched. 60mg QD组 vs 安慰剂：FVC改善差值约+160ml（达到MCID标准，即最小临床意义差异2-6%）60mg QD vs Placebo: FVC improvement difference is approx. +160ml (Reaches MCID criteria, i.e., minimal clinically important difference of 2-6%). 未使用标准治疗（SOC）患者亚组：60mg QD FVC改善 +187.8 ml（95% CI: 68.6~306.9）——效果更加显著Patient sub-group unused Standard of Care (SOC): 60mg QD FVC improved by +187.8 ml (95% CI: 68.6~306.9)—even more significant effect. 图6：Rentosertib Phase 2a临床数据（FVC变化）Figure 6: Rentosertib Phase 2a Clinical Data (Changes in FVC) 安全性：耐受良好，无严重不良事件（SAE）报告与药物相关。Safety: Well-tolerated, no Serious Adverse Events (SAE) reported related to the drug.文献： Nature Medicine, Vol. 31, pp. 2602-2610, 2025. DOI: 10.1038/s41591-025-03743-2Reference: Nature Medicine, Vol. 31, pp. 2602-2610, 2025. DOI: 10.1038/s41591-025-03743-2 后续计划：Subsequent Plans:· Phase 2b/3：全球多中心（含中国），计划2026年H1-H2启动· Phase 2b/3: Global multicenter (including China), planned to start in 2026 H1-H2.· 与主要大型药企的合作/授权谈判持续进行· Licensing/collaboration negotiations with major big pharma are ongoing. 3.2.2 RLY-2608（Relay Therapeutics）——最晚期AI药物3.2.2 RLY-2608 (Relay Therapeutics) — The Most Advanced Late-Stage AI Drug 属性 / Attribute 详情 / Details 开发公司 / Developing Company Relay Therapeutics（RLAY，纳斯达克） 靶点 / Target PI3Kα（全突变选择性抑制剂）/ PI3Kα   (Pan-mutant selective inhibitor) 适应症 / Indication PI3Kα突变的HR+/HER2-转移性乳腺癌 / PI3Kα   mutant HR+/HER2- metastatic breast cancer 阶段 / Phase Phase II已完成，Phase III计划2026年H2启动 / Phase II   completed, Phase III planned for 2026 H2 关键数据 / Key Data 无进展生存期（PFS）：9.2个月（PIK3CA突变乳腺癌）/   Progression-free survival (PFS): 9.2 months (PIK3CA mutated breast cancer) 技术特点 / Technical Features 首个口服泛突变选择性PI3Kα抑制剂（克服胰岛素代谢副作用）/   First oral pan-mutant selective PI3Kα inhibitor (overcomes insulin metabolic   side effects) 设计方法 / Design Method 结合分子动力学、冷冻电镜和DNA编码文库 / Blends   molecular dynamics, cryo-EM, and DNA-encoded libraries 发表 / Publication Cancer Discovery（背靠背论文）/ Cancer   Discovery (back-to-back papers) 投资意义：如Phase III成功，RLY-2608将成为全球首个明确证明"AI设计优于传统设计"的临床批准药物（其Pan-突变选择性是传统化学方法难以实现的）。Investment Significance:If Phase III proves successful, RLY-2608 will become the world’s first clinically approved drug to definitively prove that “AI design is superior to traditional design” (its pan-mutant selectivity is exceedingly difficult to achieve via traditional chemistry). 3.2.3 REC-4881（Recursion）——罕见病突破口3.2.3 REC-4881 (Recursion) — Breakthrough in Rare Diseases 属性 / Attribute 详情 / Details 靶点 / Target MEK1/2抑制剂 / MEK1/2   inhibitor 适应症 / Indication 家族性腺瘤性息肉病（FAP）——无已获批治疗方案的罕见病 /   Familial adenomatous polyposis (FAP) — rare disease with no approved   therapies 阶段 / Phase Phase 2（TUPELO研究）/ Phase 2   (TUPELO study) 疗效数据 / Efficacy Data 75%患者总息肉负担减少；43%中位减少（12周）；82%在25周维持 / Total polyp   burden reduced in 75% of patients; median reduction of 43% (at 12 weeks); 82%   maintained at 25 weeks 2026年计划 / 2026 Plan H1启动FDA注册路径磋商；扩展18+岁患者群体 / H1   initiation for FDA registry path consultations; expanding the 18+ patient   cohort 战略意义：FAP是罕见病，FDA孤儿药认定可加快审评；此适应症无竞争对手，若获批将是"空白市场"。Recursion Phenomics平台（细胞图像AI）在识别FAP这一非经典MEK靶点上发挥核心作用。Strategic Significance: FAP is a rare disease, and FDA orphan drug designation expedites review; there are no competitors for this indication, making it a “blank market” if approved. Recursion’s Phenomics platform (cell imaging AI) played a core role in identifying this non-classical MEK target for FAP. 3.2.4 Zasocentinib（TAK-279，Schrödinger/Takeda）——Phase III进行中3.2.4 Zasocentinib (TAK-279, Schrödinger/Takeda) — Phase III Ongoing 属性 / Attribute 详情 / Details 开发公司 / Developing Company Schrödinger（发现）+ 武田制药（Takeda，开发） / Schrödinger   (Discovery) + Takeda (Development) 靶点 / Target TYK2（酪氨酸激酶2）抑制剂 / TYK2   (Tyrosine kinase 2) inhibitor 适应症 / Indication 溃疡性结肠炎（UC）、克罗恩病（CD）/ Ulcerative   colitis (UC), Crohn’s disease (CD) 阶段 / Phase Phase   III进行中 / Phase III Ongoing 设计方法 / Design Method Schrödinger物理驱动FEP（自由能微扰）+ 机器学习 / Schrödinger   Physics-driven FEP (Free Energy Perturbation) + Machine Learning 行业地位：与RLY-2608并列为进展最晚期的AI辅助设计药物之一。TYK2靶点已被BMS的Zeposia（ozanimod）和辉瑞的Cibinqo（abrocitinib）证实在IBD领域可行，Zasocentinib代表Schrödinger技术的"最好试金石"。Industry Position: Paired with RLY-2608 as one of the most advanced late-stage AI-assisted designed drugs. The TYK2 target has been proven viable in the IBD field by BMS’s Zeposia (ozanimod) and Pfizer’s Cibinqo (abrocitinib); Zasocentinib represents the “ultimate touchstone” for Schrödinger’s technology.3.3 完整临床管线汇总3.3 Complete Clinical Pipeline Summary 药物名称 / Drug Name 公司 / Company 靶点 / Target 适应症 / Indication 阶段 / Phase 状态 / Status Zasocentinib Schrödinger/Takeda TYK2 IBD Phase   III 活跃 / Active RLY-2608 Relay PI3Kα 乳腺癌 / Breast Cancer Phase   III计划 / Phase III Planned 积极 / Positive Rentosertib Insilico TNIK IPF Phase   2a ✓ 规划Phase 2b/3 / Phase 2b/3 Planned REC-4881 Recursion MEK1/2 FAP Phase   2 活跃 / Active REC-994 Recursion — 脑海绵状血管 / Cerebral Cavernous Malformation Phase   2 去优先级 / Deprioritized RLY-1971 Relay SHP2 实体瘤 / Solid Tumors Phase   II 活跃 / Active RLY-5836 Relay FGFR2 胆管癌 / Cholangiocarcinoma Phase   II 活跃 / Active EXS-4318 Recursion CDK7 肿瘤 / Oncology Phase   II 活跃 / Active MEN2501 Insilico→Menarini 未披露 / Undisclosed 实体瘤 / Solid Tumors Phase   I ✓ 2026.02里程碑 / Feb 2026   Milestone MEN2312 Insilico→Menarini KAT6 乳腺癌 / Breast Cancer Phase   I 活跃 / Active REC-617 Recursion CDK7 卵巢癌 / Ovarian Cancer Phase   1/2 活跃 / Active REC-1245 Recursion RBM39 肿瘤 / Oncology Phase   1 活跃 / Active ISM5411 Insilico 未披露 / Undisclosed SLE（红斑狼疮）/ SLE (Lupus) Phase   1/2 活跃 / Active ISM3312 Insilico 3CLpro COVID-19 Phase   I 活跃 / Active EXS21546 Recursion A2a 肿瘤免疫 / Immuno-oncology Phase   I（2020年启 / Started 2020） 活跃 / Active DSP-0038 Recursion 5-HT1a/2a 阿尔茨海默症精神病 / Alzheimer’s Psychosis Phase   I 活跃 / Active SGR-1505 Schrödinger MALT1 血液肿瘤 / Hematologic Malignancies Phase   I 活跃 / Active SGR-2921 Schrödinger CDC7 实体瘤 / Solid Tumors Phase   I 活跃 / Active BEN-8744 BenevolentAI PDE10A 溃疡性结肠炎 / Ulcerative Colitis Phase   I/II 数据积极 / Data Positive DSP-1181 Exscientia（历史 / Historical） 5-HT1A OCD Phase   I（2020年 / 2020） 已停止 / Halted 来源：Axis Intelligence 2026年分析报告；公司IR公告；Nature Medicine；ClinicalTrials.govSource: Axis Intelligence 2026 Analysis Report; Company IR Filings; Nature Medicine; ClinicalTrials.gov 第四章：七大技术路线深度分析------------------Chapter 4: In-depth Analysis of Seven Technical Routes4.1 技术路线总图4.1 Tech Route Overview Map AI制药技术可分为七大核心路线，覆盖药物发现全链条：AIDD technology can be broken down into seven core routes, covering the entire drug discovery chain: ↑ ↑ ↑ ↑ ↑知识图谱生成对抗网络强化学习深度学习因果AIKnowledge Graph, GAN, Reinforcement Learning, Deep Learning, Causal AINLP/LLM 扩散模型贝叶斯优化 GNN 数字孪生NLP/LLM, Diffusion Models, Bayesian Optimization, GNN, Digital Twins多组学AI VAE/Transformer 多目标优化物理计Multi-omics AI, VAE/Transformer, Multi-objective Optimization, Physics Computation4.2 技术路线一：蛋白质结构预测（最成熟，★★★★★）4.2 Technical Route One: Protein Structure Prediction (Most Mature, ★★★★★) 蛋白质结构预测是AI制药技术革命的"原点"。2024年诺贝尔化学奖颁发给David Baker（蛋白质从头设计）、Demis Hassabis和John Jumper（AlphaFold），是对这一技术路线最高的学术背书。Protein structure prediction is the “origin” of the AI drug discovery technological revolution. The 2024 Nobel Prize in Chemistry awarded to David Baker (de novo protein design), Demis Hassabis, and John Jumper (AlphaFold) is the highest academic endorsement of this technical route. AlphaFold 2（2020年）：在CASP14竞赛中以压倒性优势超越所有传统方法，蛋白质结构预测精度首次达到实验级别（原子级分辨率）。已预测2.14亿个蛋白质结构（几乎覆盖所有已知蛋白质组），公开发布的AlphaFold数据库成为全球研究人员的必备工具。AlphaFold 2 (2020):Overwhelmingly surpassed all traditional methods in the CASP14 competition, attaining experimental-level accuracy (atomic resolution) for protein structure prediction for the first time. It has predicted 214 million protein structures (covering almost all known proteomes), and the publicly released AlphaFold database has become an essential tool for global researchers. AlphaFold 3（2024年）：扩展至预测蛋白质与小分子、核酸、蛋白质之间的相互作用（复合物结构），使虚拟筛选精度大幅提升。技术架构引入扩散模型（Diffusion Model），结合Evoformer进化序列信息。AlphaFold 3 (2024): Expanded to predict interactions between proteins and small molecules, nucleic acids, and other proteins (complex structures), vastly improving virtual screening precision. The technical architecture introduced Diffusion Models, combined with Evoformer evolutionary sequence data. "AlphaFold 4"级别新模型（2026年初）：Nature报道Isomorphic Labs发布下一代模型，据称在动态构象预测和诱导契合效应上实现进一步突破，为AI药物设计提供更精确的结构基础。“AlphaFold 4” level new models (Early 2026): Nature reports that Isomorphic Labs released next-generation models, allegedly achieving further breakthroughs in dynamic conformation prediction and induced-fit effects, providing more precise structural foundations for AI drug design. 代表应用：Representative Applications:- Isomorphic Labs（基于AlphaFold，已开始人体试验）- Isomorphic Labs (based on AlphaFold, human trials initiated)- Xaira Therapeutics（David Baker团队RFdiffusion）- Xaira Therapeutics (David Baker’s team, RFdiffusion)- 全球前20大药企均已将AlphaFold整合入药物发现管线- The world’s top 20 pharmaceutical companies have all integrated AlphaFold into their drug discovery pipelines4.3 技术路线二：生成式AI分子设计（★★★★☆）4.3 Technical Route Two: Generative AI Molecular Design (★★★★☆) 小分子生成AI是产生临床候选药物最直接的技术路线，经历了从GAN→VAE→Transformer→扩散模型的快速迭代。Small molecule generation AI is the most direct technical route to producing clinical candidate drugs, going through rapid iterations from GANs → VAEs → Transformers → Diffusion Models. 生成对抗网络（GAN）：早期主流方法，Insilico早期的GENTRL模型；问题：训练不稳定，生成分子可合成性较差。Generative Adversarial Networks (GANs): Early mainstream method, such as Insilico’s early GENTRL model; Issues: Unstable training, poor synthesizability of generated molecules. 变分自编码器（VAE）：将分子编码到连续潜空间，支持梯度优化；Insilico的Chemistry42早期版本采用此方案。Variational Autoencoders (VAEs): Encodes molecules into a continuous latent space, supporting gradient optimization; Insilico’s early Chemistry42 version applied this approach. Transformer分子语言模型：将SMILES字符串视为"语言"，大规模预训练后fine-tune；Rentosertib即是通过此路线发现的TNIK抑制剂骨架。代表模型：ChemBERTa、MolGPT、Chemistry42新版。Transformer Molecular Language Models: Treats SMILES strings as “language,” applying fine-tuning after large-scale pre-training; Rentosertib’s TNIK inhibitor scaffold was discovered via this route. Representative models: ChemBERTa, MolGPT, Chemistry42 (newer version). 扩散模型（当前最前沿）： DiffSBDD、Pocket2Mol等模型基于蛋白质口袋结构直接生成靶向小分子；Isomorphic Labs下一代引擎采用此路线；生成质量和多样性显著优于前代。Diffusion Models (current frontier): Models like DiffSBDD and Pocket2Mol directly generate targeted small molecules based on protein pocket structures; Isomorphic Labs’ next-gen engine employs this route; generation quality and diversity are significantly superior to the previous generations.4.4 技术路线三：强化学习分子优化（★★★★☆）4.4 Technical Route Three: Reinforcement Learning for Molecular Optimization (★★★★☆) 强化学习（RL）将分子优化问题建模为序贯决策过程，奖励函数由多目标属性（活性、选择性、类药性、合成可行性等）组合定义。Reinforcement Learning (RL) models the molecular optimization problem as a sequential decision process, wherein the reward function is defined by a combination of multi-objective attributes (activity, selectivity, drug-likeness, synthetic feasibility, etc.). 代表应用：Representative Applications:- Chemistry42（Insilico）：RL + 生成模型组合，同时优化多个化学属性- Chemistry42 (Insilico): RL + Generative model combinations, simultaneously optimizing multiple chemical parameters.- Recursion OS：在Phenomics数据引导下进行化合物空间探索- Recursion OS: Exploring compound space under the direction of Phenomics data.- Exscientia（历史）Centaur Chemist：AI驱动的化学空间高效探索，DSP-1181仅合成380个化合物完成Phase 1候选发现（传统需数千个）- Exscientia (historical) Centaur Chemist: AI-driven highly efficient exploration of chemical space; DSP-1181 synthesized only 380 compounds to complete Phase 1 candidate discovery (traditional methods require thousands).4.5技术路线四：图神经网络与ADMET预测（★★★★☆）4.5 Technical Route Four: Graph Neural Networks & ADMET Prediction (★★★★☆) 图神经网络（GNN）天然适配分子图结构，是ADMET（吸收、分布、代谢、排泄、毒性）预测的主流技术。Graph Neural Networks (GNN) naturally suit molecular graph structures and are the mainstream technology for ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) predictions. 核心价值：在合成候选分子之前预测其类药性，大幅减少失败率高的化合物进入合成阶段。Core Value: Predicting drug-likeness of candidate molecules before synthesizing them significantly reduces high-failure-rate compounds from entering the synthesis phase. 代表技术：Representative Technologies:- Schrödinger的FEP+（自由能微扰，物理驱动+ML混合方法）：Zasocentinib开发关键工具- Schrödinger’s FEP+ (Free Energy Perturbation, a physics-driven + ML hybrid method): A key tool in Zasocentinib development.- MIT的Graph Networks for Materials Exploration- MIT’s Graph Networks for Materials Exploration.- Atomwise的AtomNet（深度卷积神经网络，用于虚拟筛选）- Atomwise’s AtomNet (Deep Convolutional Neural Network used for virtual screening). 局限性：ADMET预测精度仍不完善，特别是代谢预测和毒性预测；多个候选药物仍因实际毒性与预测不符而终止。Limitations: ADMET prediction accuracy remains imperfect, especially for metabolism and toxicity predictions; several candidate drugs have still been terminated because actual toxicity misaligned with predictions.4.6技术路线五：大语言模型（LLM）应用（★★☆☆☆，快速成长中）4.6 Technical Route Five: LLM Large Language Model Applications (★★☆☆☆, Fast-growing) LLM在制药中的应用处于早期阶段，但潜力巨大：LLM applications in pharmaceuticals remain early-stage but hold immense potential: 文献挖掘与靶点识别：训练于PubMed/bioRxiv的专业模型（BioGPT、PubMedGPT）能从千万篇论文中提取靶点-疾病关联信息，是BenevolentAI和Recursion靶点识别的重要工具。Literature Mining and Target Identification: Specialized models trained on PubMed/bioRxiv (BioGPT, PubMedGPT) can extract target-disease association information from tens of millions of papers, acting as vital target identification tools for BenevolentAI and Recursion. 多模态临床数据分析：整合电子病历、基因组、蛋白组数据，辅助患者分层和生物标志物发现。Multimodal Clinical Data Analysis: integrating electronic medical records, genomics, and proteomics data to aid patient stratification and biomarker discovery. 药物说明书与监管文件生成：辅助加速监管申请材料撰写（FDA正在评估AI生成监管文件的政策）。Drug Labels and Regulatory Document Generation: Assisting in accelerating the composition of regulatory application materials (the FDA is assessing non-binding policies for AI-generated regulatory documents). AI Agent实验规划（最前沿）：自主规划实验方案，结合机器人实验室（Emerald Cloud Lab、Strateos）实现"AI科学家"闭环；预期2026-2028年进入早期应用。AI Agent Experimental Planning (Frontier): Autonomous planning of experimental protocols, combining robotic labs (Emerald Cloud Lab, Strateos) to manifest an “AI Scientist” loop; early applications expected between 2026-2028.4.7 技术路线六：表型AI——Recursion模式（★★★☆☆）4.7 Technical Route Six: Phenotypic AI — The Recursion Model (★★★☆☆) 表型AI（Phenotypic AI）是Recursion独特的技术路线：不从已知靶点出发，而是从细胞形态学变化（表型）反推药物机制。Phenotypic AI is Recursion’s unique trajectory: deviating away from known targets, it inversely deduces drug mechanisms derived from changes in cell morphology (phenotypes). 核心方法：Core Methods:- 高内涵成像（High-Content Imaging）：每周拍摄数百万张细胞图像- High-Content Imaging: Taking millions of cell images weekly.- CNN特征提取：从图像中提取数千个细胞形态特征- CNN Feature Extraction: Pulling thousands of cellular morphological features from images.- 表型图谱（Phenomaps）：构建基因-化合物-细胞形态三维关联网络- Phenomaps: Building 3D associative networks covering gene-compound-cell morphology. 代表成果：Representative Outcomes:- Neuron Map（1万亿iPSC来源神经元全基因组CRISPR图谱）：Roche/Genentech支付3,000万美元- Neuron Map (1 trillion iPSC-derived neuron whole-genome CRISPR map): Roche/Genentech paid $30 million.- REC-4881在FAP的应用：Phenomics平台识别出MEK作为FAP的新治疗靶点（此前非公认标准靶点）- REC-4881 in FAP application: Phenomics platform identified MEK as a novel therapeutic target for FAP (not previously recognized as a standard target). 局限性：从表型反推靶点存在机制不明确的风险；数据量庞大但解读挑战大；成本高。Limitations: Progressing inversely from phenotype to target involves risks of unclear mechanisms; although the data volume is huge, interpreting it is a monumental challenge; high costs.4.8 技术路线七：蛋白质从头设计（★★★☆☆，高潜力）4.8 Technical Route Seven: De Novo Protein Design (★★★☆☆, High Potential) 蛋白质从头设计（de novo protein design）是最前沿的技术路线，由2024年诺贝尔奖得主David Baker团队主导，Xaira Therapeutics是最大的商业化推手。De novo protein design defines the most frontier tech route, spearheaded by 2024 Nobel Laureate David Baker’s team, with Xaira Therapeutics pushing heavily on the commercialization front. RFdiffusion（2023年）：基于扩散模型的蛋白质骨架设计，可生成针对特定功能的全新蛋白质；已设计出多种结合特定小分子的蛋白质和蛋白质-蛋白质结合界面。RFdiffusion (2023):Based on diffusion-model protein scaffold layouts to generate wholly novel feature-specific proteins; it has successfully designed multiple distinct proteins correlating heavily with specified small molecules and protein-protein binding interfaces. 药物应用：Drug Applications:- 全新蛋白质药物（不依赖抗体、多肽等已有骨架）- Novel protein drugs (completely independent from existing antibody or polypeptide scaffolding).- 靶向"难成药"蛋白-蛋白相互作用（PPI）靶点- Targeting “undruggable” protein-protein interactions (PPI) targets.- 双特异性和多特异性蛋白质设计- Bispecific and multispecific protein design.- 分子胶（Molecular Glue）和PROTAC靶向蛋白降解- Molecular Glues and PROTAC targeted protein degradation. 当前阶段：绝大多数应用仍在临床前阶段；Xaira Therapeutics以超10亿美元A轮融资推进该路线，预期2027-2030年出现首个临床候选。Current Phase: The vast majority of applications are strictly traversing preclinical phases; Xaira Therapeutics pushes this route with its >$1 billion Series A financing, expecting an initial clinical candidate emerging by 2027-2030.4.9 各公司技术路线对比4.9 Comparison of Tech Routes Across Companies 公司 / Company 核心技术路线 / Core Tech    Route 专有模型/平台 / Proprietary Model/Platform 特色优势 / Competitive    Advantage 临床阶段 / Clinical    Phase Insilico Medicine 生成AI（Transformer+RL）+ 蛋白质结构 / Generative   AI (Transformer+RL) + Protein Structure Pharma.AI（Biology42+Chemistry42） 端到端；首个Phase 2a RCT验证 / End-to-end;   First Phase 2a RCT Validated Phase 2a ✓ Recursion 表型ML + 生物图谱 / Phenotypic   ML + Biol-Maps Recursion OS；Phenomaps 超大规模表型数据（细胞图像）/   Ultra-large scale phenotypic data (cell images) Phase 2 Isomorphic Labs AlphaFold + 扩散模型 / AlphaFold +   Diffusion Models 下一代药物设计引擎 /   Next-gen drug design engine 蛋白质结构领先；Alphabet资源 / Leader in   protein structure; Alphabet resources 人体试验启动 / Human   trials initiated Relay Therapeutics 分子动力学 + 冷冻电镜 + DNA文库 / Molecular   Dynamics + Cryo-EM + DNA Libraries Dynamo平台 / Dynamo   Platform 构象灵活性优化 /   Conformational flexibility optimization Phase III计划 / Phase III   Planned Xaira Therapeutics 蛋白质从头设计（RFdiffusion）/ De novo protein   design (RFdiffusion) 未公开 / Undisclosed David Baker团队；超10亿美元资本 / David Baker   team; Over $1B capital 临床前 / Preclinical BenevolentAI NLP知识图谱 + GNN / NLP   Knowledge Graph + GNN BenevolentAI Platform 文献靶点挖掘 / Literature   mining for targets Phase I/II Schrödinger 物理力场 + ML混合 / Physics force   field + ML hybrid FEP+, WaterMap 高精度计算；软件+管线双轮 /   High-precision computation; Dual software+pipeline wheel Phase I/III（合作 / Collaborated） 晶泰科技 / XtalPi 量子力学 + ML晶型预测 / Quantum   mechanics + ML polymorph prediction 晶型预测引擎 / Polymorph   prediction engine 固态药物形式专家 /   Solid-state drug format expert 服务模式 / Service   Model 图7：AI制药技术路线成熟度评估Figure 7: Evaluation of AI Drug Discovery Tech Route Maturity 图8：各公司技术路线矩阵对比Figure 8: Comparative Matrix of Tech Routes Across Companies -–第五章：核心公司深度分析Chapter 5: In-Depth Analysis of Core Companies5.1 Insilico Medicine（英矽智能，3696.HK）——行业标杆5.1 Insilico Medicine (3696.HK) — Industry Benchmark公司概况Company Overview 属性 / Attribute 详情 / Details 成立时间 / Founded 2014年 / 2014 总部 / Headquarters 香港（主要运营在中国大陆和美国）/ Hong Kong (primary operations in mainland China and the US) 创始人 / Founder Alex Zhavoronkov（CEO） 股票代码 / Ticker 3696.HK（香港联交所主板） / 3696.HK (HKEX Main Board) IPO日期 / IPO Date 2025年12月30日 / December 30, 2025 IPO募资 / IPO Raised 22.77亿港元（约2.9亿美元）/ HKD 2.277 Billion (~$290 million) IPO超额认购 / IPO Oversubscription 公开部分1,427.37倍（历史罕见）/ Public tranche 1,427.37x (historically rare) 市场地位 / Market Position AI制药领域最具临床验证的纯AI原生公司 / The most clinically verified pure AI-native company in AI drug   discoveryPharma.AI平台架构Pharma.AI Platform Architecture Insilico的核心竞争力是端到端AI药物发现平台Pharma.AI，涵盖四个模块：Insilico’s core competency lies in its end-to-end AI drug discovery platform Pharma.AI, which spans four modules:- Biology42：靶点识别与验证（从多组学数据挖掘疾病相关靶点）- Biology42: Target identification and validation (mining disease-associated targets from multi-omics data).- Chemistry42：AI驱动小分子生成与优化（超过300个已知生成模型）- Chemistry42: AI-driven small molecule generation and optimization (over 300 known generative models).- Medicine42：临床结果预测（患者分层、临床试验设计优化）- Medicine42: Clinical outcome predictions (patient stratification, clinical trial design optimization).- Science42：综合科学知识管理与推理- Science42: Comprehensive scientific knowledge management and reasoning. 平台商业化：全球前20大制药公司中13家已授权Pharma.AI软件；3项资产已对外授权，合约总价值超20亿美元。Platform Commercialization: 13 of the top 20 global pharma companies have licensed Pharma.AI software; 3 assets have been out-licensed, with total contract values exceeding $2 billion.财务数据Financial Data 年份 / Year 营收（百万美元）/ Revenue (USD M) 净亏损（百万美元）/ Net Loss (USD M) 2022 30.1 222 2023 51.2 212 2024 85.8 17 2025 H1 27.5 19 注意：2024年净亏损大幅收窄主要因公允价值变动收益（非经常性），投资者需关注经调整净亏损。专利及专利申请达787项。Note: The sharp narrowing of the net loss in 2024 is mainly attributed to fair value adjustment gains (non-recurring). Investors should focus on adjusted net loss. Patents and applications total 787.临床管线Clinical Pipeline 药物 / Drug 靶点 / Target 适应症 / Indication 阶段 / Phase 最新进展 / Latest Progress Rentosertib（ISM001-055） TNIK 特发性肺纤维化（IPF） / IPF Phase 2a ✓ Nature Medicine发表；规划全球Phase 2b/3 / Published in Nature Medicine; Global Phase 2b/3   planned ISM5411 未披露 / Undisclosed 系统性红斑狼疮（SLE） / SLE Phase 1/2 进行中 / Ongoing MEN2501（授权给Menarini） / (Licensed to Menarini) 未披露 / Undisclosed 实体瘤 / Solid tumors Phase I ✓ 2026年2月获500万美元里程碑 / Received $5M milestone in Feb 2026 MEN2312（授权给Menarini） / (Licensed to Menarini) KAT6 乳腺癌 / Breast cancer Phase I 进行中 / Ongoing 20+其他项目 / 20+ other projects 多靶点 / Multi-target 多适应症 / Multi-indication IND/临床前 / IND/Pre-clinical — 投资亮点：Investment Highlights:- 首个完成RCT的AI设计药物（Rentosertib），临床效力已证明- First AI-designed drug to complete an RCT (Rentosertib), showing proven clinical efficacy.- 软件授权收入可提供相对稳定的非管线收入- Software licensing revenue provides relatively stable non-pipeline income.- 港股上市后进入公开市场，流动性显著提升- Entering the public market via HKEX significantly improves liquidity.- 基石投资者（礼来、腾讯、淡马锡）背书强- Strongly endorsed by cornerstone investors (Eli Lilly, Tencent, Temasek). 主要风险：Major Risks:- Rentosertib Phase 2b/3是最大价值催化剂，失败将大幅打击估值- Rentosertib Phase 2b/3 is the maximum-value catalyst; failure would devastate valuation.- 整体仍处亏损状态，现金消耗需持续融资支持- Overall still operating at a loss; cash burn necessitates continuous financing support.- 中国市场监管合规风险（数据安全、香港上市规则）- Regulatory compliance risks in the Chinese market (Data safety, HK listing rules). -–5.2 Recursion Pharmaceuticals（RXRX）——TechBio整合者5.2 Recursion Pharmaceuticals (RXRX) — TechBio Integrator公司概况Company Profile 属性 / Attribute 详情 / Details 成立时间 / Founded 2013年 / 2013 总部 / Headquarters 美国犹他州盐湖城 / Salt Lake City, Utah, USA 股票代码 / Ticker RXRX（纳斯达克） / RXRX (Nasdaq) 关键事件 / Key Event 2024年11月完成与Exscientia合并 / Completed merger with Exscientia in Nov 2024 市场定位 / Market Positioning “TechBio”——技术驱动的生物制药公司 / “TechBio” — A technology-driven biopharmaceutical company2025年全年财务数据（2026年2月25日公告）Full Year 2025 Financial Data (Announced Feb 25, 2026) 指标 / Metric 2025全年 / 2025 FY 2024全年 / 2024 FY 变化 / Change 总营收 / Total Revenue 7,470万美元 / $74.7M 5,880万美元 / $58.8M +27% R&D支出 / R&D   Expenses 4.753亿美元 / $475.3M 3.144亿美元 / $314.4M +51% 净亏损 / Net Loss 6.448亿美元 / $644.8M 4.637亿美元 / $463.7M +39% 现金及等价物 / Cash &   Equivalents 7.539亿美元 / $753.9M 6.030亿美元 / $603.0M +25% 现金runway预期 / Cash Runway   Expectation 至2028年初 / To early 2028 — — 2026年运营支出指引 / 2026 OpEx   Guidance <3.90亿美元 / <$390M — 主动缩减 / Proactive   reduction合作收入（已验证数据）Collaboration Revenues (Validated Data) 合作方 / Partner 项目内容 / Project Scope 已收取里程碑 / Milestones Collected Sanofi 5+个AI驱动小分子项目（免疫学+肿瘤学） 5+ AI-driven small molecule   projects (Immunology+Oncology) 1.34亿美元（共5次里程碑）$134 million (Total of 5   milestones) Roche/Genentech Phenomap数据合作（神经元图谱+微胶质细胞图谱） Phenomap data collab (Neuron   map+Microglia map) 2.13亿美元（预付款+里程碑）$213 million   (Upfront+Milestones) 合计 / Total — >3.47亿美元 / >$347   million临床管线（五个差异化项目）Clinical Pipeline (Five Differentiated Projects) 药物 / Drug 靶点 / Target 适应症 / Indication 阶段 / Phase 2026年计划 / 2026 Plan REC-4881 MEK1/2 FAP（家族性腺瘤性息肉病） / FAP Phase 2 ✓ H1 FDA注册路径磋商 / H1 FDA   registry path consultations REC-617 CDK7 铂耐药卵巢癌 /   Platinum-resistant ovarian cancer Phase 1/2 — REC-1245 RBM39降解剂 / RBM39   degrader 肿瘤 / Oncology Phase 1 H1安全性/PK数据 / H1 Safety/PK   data REC-3565 MALT1 — Phase 1 — EXS-4318 CDK7 肿瘤 / Oncology Phase 2 — 注：合并Exscientia后已终止REC-2282（NF2）、REC-994（CCM）、REC-3964共三个项目，集中资源在差异化项目上。Note: Post-Exscientia merger, three projects—REC-2282 (NF2), REC-994 (CCM), REC-3964—were terminated to concentrate resources on differentiated projects. 投资亮点： Sanofi/Roche合作提供稳定里程碑收入；平台技术（Phenomics）可扩展至多个大型制药合作；REC-4881在FAP的孤儿药身份可享监管优惠。Investment Highlights: Sanofi/Roche collaborations provide stable milestone revenues; the platform tech (Phenomics) is scalable explicitly to multiple mega-pharma partnerships; REC-4881’s orphan designation in FAP enjoys regulatory advantages. 主要风险：2025年净亏损6.45亿美元且扩大；R&D投入激增但商业化仍遥远；合并Exscientia的整合复杂性；管线终止项目数量较多。Major Risks: Massive and growing 2025 net loss of $645M; soaring R&D costs with commercialization still far distant; integration complexity from merging with Exscientia; numerous terminated pipeline projects. -–5.3 Isomorphic Labs（Alphabet旗下）——结构生物学+AI的最强结合5.3 Isomorphic Labs (Alphabet) — The Strongest Fusion of Structural Biology + AI 属性 / Attribute 详情 / Details 成立时间 / Founded 2021年（从Google DeepMind分拆） / 2021 (Spun   off from Google DeepMind) 母公司 / Parent   Company Alphabet（谷歌母公司） / Alphabet   (Google parent) 总部 / Headquarters 英国伦敦 / London, UK 外部融资 / External   Funding 2025年3月完成6亿美元A轮（领投：Thrive Capital）/ Complete $600M   Series A in Mar 2025 (Lead: Thrive Capital) 合作方 / Partners Eli Lilly、Novartis（已签署合作协议）/ Eli   Lilly, Novartis (Signed agreements) 临床进展 / Clinical   Progress 2025年7月宣布首批AI设计候选进入人体试验 /   Initial AI candidates announced for human trials July 2025 核心技术壁垒 / Core Tech   Moat AlphaFold 2/3及下一代模型（全球领先的蛋白质结构预测能力）/ AlphaFold 2/3 & Next-gen models (Global leader in protein   structure prediction) 战略定位： Isomorphic Labs代表了"大科技公司"进军制药最深入的布局。其技术基础（AlphaFold）已获诺贝尔奖背书，商业化路径是将结构预测优势转化为药物设计能力。Strategic Positioning:Isomorphic Labs represents “Big Tech’s” deepest penetration into pharmaceuticals. Its foundational technology (AlphaFold) bears Nobel-prize endorsement, with a commercial pathway leveraging structural prediction advantages to manifest actual drug design capabilities. 关键不确定性：首批临床候选具体靶点/适应症未公开；临床进展透明度低于上市公司；尚无公开临床数据。Key Uncertainties: The specific targets/indications of the initial clinical candidates remain undisclosed; clinical transparency sits lower than public entities; essentially zero public clinical data currently available. -–5.4 Relay Therapeutics（RLAY）——构象生物学先驱5.4 Relay Therapeutics (RLAY) — Conformational Biology Pioneer 属性 / Attribute 详情 / Details 成立时间 / Founded 2016年 / 2016 股票代码 / Ticker RLAY（纳斯达克） / RLAY   (Nasdaq) 核心技术 / Core Tech Dynamo平台（分子动力学模拟 + 冷冻电镜 + DNA编码文库） / Dynamo   Platform (Molecular Dynamics + Cryo-EM + DNA Encoded Libraries) 核心理念 / Core Concept 靶向蛋白质的构象动态（“motion”），而非仅静态结构 /   Targeting the conformational dynamics of proteins (“motion”), instead of   purely static structures 管线亮点（RLY-2608）：Pipeline Highlights (RLY-2608):- 全突变选择性PI3Kα抑制剂，克服传统PI3Kα抑制剂的胰岛素代谢副作用- Pan-mutant selective PI3Kα inhibitor, overcoming traditional PI3Kα inhibitor’s insulin-driven metabolic side effects.- Phase II PFS 9.2个月（PIK3CA突变乳腺癌）- Phase II PFS of 9.2 months (for PIK3CA mutant breast cancer).- Phase III计划2026年H2启动- Phase III planned initiation in 2026 H2.- 与现有标准治疗（Alpelisib）相比，选择性显著更高，副作用更少- Compared with existing standard-of-care (Alpelisib), selectivity is vastly superior, presenting significantly fewer side-effects. 投资逻辑：乳腺癌PI3Kα突变靶点市场巨大；RLY-2608的差异化（无代谢毒性）是其竞争优势；Phase III结果是最重要的价值催化剂。Investment Logic: The PI3Kα mutated breast cancer target signifies a vast market; RLY-2608’s primary differentiation (free from metabolic toxicities) denotes a sharp competitive advantage; Phase III results act as the most essential value catalyst entirely. -–5.5 Xaira Therapeutics——最大A轮，最高期望5.5 Xaira Therapeutics — Largest Series A, Highest Expectations 属性 / Attribute 详情 / Details 成立时间 / Founded 2024年 / 2024 融资 / Funding 超10亿美元（2024年A轮，AI制药史上最大A轮） / Over $1   Billion (2024 Series A, largest in AI drug discovery history) 联合创始 / Co-Founder ARCH Venture Partners 技术背景 / Tech   Background David Baker团队（2024年诺贝尔化学奖，蛋白质从头设计） /   David Baker Team (2024 Nobel in Chemistry, de novo protein design) 核心技术 / Core Tech RFdiffusion蛋白质从头设计 + 生成AI / RFdiffusion   de novo protein design + GenAI 当前阶段 / Current   Phase 早期管线开发，无公开临床信息 /   Early pipeline development, no open clinical info 融资意义：超10亿美元A轮是VC对AI蛋白质设计技术的巨大赌注。David Baker团队的科学信誉是核心资产。但距临床还有数年。Funding Significance:The >$1 Billion Series A is a colossal VC gamble on AI protein design technology. David Baker’s team’s scientific credibility is its chief asset. Conversely, clincal testing remains several years out. -–5.6 Schrödinger（SDGR）——老牌计算化学+AI转型5.6 Schrödinger (SDGR) — Veteran Computational Chemistry pivoting to AI 属性 / Attribute 详情 / Details 成立时间 / Founded 1990年 / 1990 股票代码 / Ticker SDGR（纳斯达克） / SDGR   (Nasdaq) 商业模式 / Business   Model 双轮驱动：软件授权（Schrödinger Suite）+ 内部管线 / Dual-drive:   Software licensing (Schrödinger Suite) + Internal pipeline 软件年收入 / Software   Annual Rev 约2亿美元/年（相对稳定） / Approx.   $200M/year (Relatively stable) 核心技术 / Core Tech 物理驱动（分子动力学/量子力学）+ 机器学习混合方法 /   Physics-driven (Molecular Dynamics/Quantum Mechanics) + Machine Learning   hybrid algorithms 主要合作 / Key Partners Bill Gates、Pfizer、Sanofi、BMS 内部管线亮点：Internal Pipeline Highlights:- SGR-1505（MALT1抑制剂）：血液肿瘤，Phase I- SGR-1505 (MALT1 inhibitor): Hematologic malignancies, Phase I- SGR-2921（CDC7抑制剂）：实体瘤，Phase I- SGR-2921 (CDC7 inhibitor): Solid tumors, Phase I- Zasocentinib（TAK-279，授权给Takeda）：TYK2，IBD，Phase III（进展最晚期的Schrödinger相关资产）- Zasocentinib (TAK-279, licensed to Takeda): TYK2, IBD, Phase III (the most advanced Schrödinger-associated asset) 投资逻辑：软件收入提供稳定基础；内部管线提供上行期权；Zasocentinib Phase III结果是近期最重要催化剂。Investment Logic: Software revenue provides a stable base framework; internal pipelines deliver powerful upside optionality; Zasocentinib Phase III results characterize the most critical near-term catalyst. -–5.7 BenevolentAI——警示案例5.7 BenevolentAI — A Cautionary Tale BenevolentAI是AI制药行业最典型的"高开低走"案例，对行业估值定价具有重要参考意义。BenevolentAI acts as the archetypal “started high, crashed low” benchmark across AI drug discovery. It carries profound referential value for industry valuation metrics. 事件 / Event 时间 / Time 影响 / Impact 阿姆斯特丹上市 / Amsterdam Listing 2022年 / 2022 估值超10亿英镑 / Valued largely over £1 Billion 核心资产临床失败 / Core Asset Clinical Failure 2023年 / 2023 股价暴跌 / Stock plummeted 大规模重组、裁员、关闭美国办公室 / Massive   restructuring/layoffs, US office closed 2024年 / 2024 市值仅剩约8,500万英镑（-90%+） / Market cap collapsed to merely ~£85   Million (-90%+) 保留BEN-8744（UC管线）、专注核心 / Retained BEN-8744 (UC   pipeline), focused on core 2024-2025年 / 2024-2025 收缩策略 / Contraction strategy 教训：Lessons Learned:1. 临床失败的惩罚极其严重——单个关键资产失败可导致90%+估值损失 1.The penalty for clinical failure is astronomically severe—a single distinct clinical failure destroys 90%+ in valuation losses.2. 平台技术商业化路径需有临床验证支撑，纯技术概念难以长期支撑估值 2.Platform tech commercialization pathways necessitate clinical verifications; pure technological concepts struggle significantly to uphold valuation.3. SaaS业务（知识探索工具）无法在临床失败时对冲估值风险 3.SaaS businesses (knowledge exploration frameworks) absolutely cannot hedge valuations actively during crushing clinical failures. -–5.8 晶泰科技（XtalPi，2228.HK）——中国固态药物AI专家5.8 XtalPi (2228.HK) — China’s AI Solid-state Drug Expert 属性 / Attribute 详情 / Details 成立时间 / Founded 2015年（麻省理工学院衍生） / 2015   (MIT spin-off) 总部 / Headquarters 深圳 / Shenzhen 上市 / Listed 2024年7月，香港主板（2228.HK） / July 2024, HKEX   Main Board (2228.HK) 技术专长 / Technical   Forte 量子力学 + AI晶型预测、固态药物形式筛选 /   Quantum Mechanics + AI polymorph predictions, solid-state formulation   screening 主要客户 / Major   Clients 辉瑞、强生、礼来等全球前20大药企 / Pfizer,   Johnson & Johnson, Eli Lilly, etc. 商业模式 / Business   Model 技术服务（CRO）为主 + 少量自有管线 /   Predominantly technical services (CRO) + handful of internal pipeline assets 定位差异化：晶泰专注于药物固态形式（多晶型）预测，这是传统药物开发的关键环节但长期依赖实验。AI预测显著降低晶型筛选成本和时间，服务黏性强。港股上市后正扩展至AI赋能CDMO业务。Positioning Differentiation: XtalPi expressly focuses heavily on tracking solid-state formats (polymorph) predictions, a perpetually critical hub intrinsically bound to lengthy experimentation phases. AI prediction observably reduces screening times and expenses; engendering tremendous service retention. Listing on HKEX facilitates expansions scaling strictly into AI-enabled CDMO business parameters. -– 图9：主要公司现金储备与runway对比Figure 9: Comparison of Main Companies’ Cash Reserves and Runway 图10：主要公司融资规模对比Figure 10: Comparison of Financing Scale Among Major Companies -–第六章：资本市场与融资格局Chapter 6: Capital Markets and Financing Landscape6.1 全球AI制药投融资趋势6.1 Global AI Drug Discovery Investment and Financing Trends 年份 / Year VC投资额 / VC    Investment 同比变化 / YoY Change 备注 / Remarks 2023 28亿美元 / $2.8B — 疫情后回调，整体生物科技低迷 /   Post-pandemic retracement, bleak overarching biotechnology context 2024 32亿美元 / $3.2B +14% 企稳回升 / Stabilizing   and recovering 2025 57亿美元 / $5.7B +78% Isomorphic融资带动，头部公司超大额融资 /   Bolstered by Isomorphic funding, gargantuan scale fundings via top-tier   groups 2026E 72-88亿美元 / $7.2-8.8B +26-54% 临床验证推动持续增长 /   Clinical validations powering continual surging growth 图11：2023-2026年AI制药投融资趋势Figure 11: AI Drug Discovery Investment Trends (2023-2026)6.2 重大融资事件6.2 Major Financing Events Xaira Therapeutics（2024年）：超10亿美元A轮Xaira Therapeutics (2024): Over $1 Billion Series A- AI制药史上最大A轮融资，由ARCH Venture Partners联合创立- The largest Series A financing round fundamentally spanning AI drug discovery histories, co-founded directly by ARCH Venture Partners.- 背书：David Baker 2024年诺贝尔化学奖- Endorsement: David Baker, 2024 Nobel Prize winner in Chemistry. Isomorphic Labs（2025年3月）：6亿美元A轮Isomorphic Labs (March 2025): $600 Million Series A- 首次引入外部资本（此前完全由Alphabet资助）- Integrating external capital channels for the first time natively (previously wholly funded structurally by Alphabet entirely).- 领投：Thrive Capital；跟投：GV、Alphabet- Lead: Thrive Capital; Following entities: GV, Alphabet. 英矽智能IPO（2025年12月）：港交所主板，募集22.77亿港元Insilico Medicine IPO (Dec 2025): HKEX Main Board, raising HKD 2.277 Billion- 2025年港股最大生物科技IPO- The comprehensively largest biotech IPO across Hong Kong markets tracking strictly effectively in 2025.- 公开发售超额认购1,427.37倍，创港股生物科技IPO历史记录- Highly historic public offering drastically oversubscribed by 1,427.37x.- 基石投资者：礼来、腾讯、淡马锡、施罗德、瑞银、橡树资本- Cornerstone investors: Eli Lilly, Tencent, Temasek, Schroders, UBS, Oaktree Capital.6.3 大型制药合作里程碑付款6.3 Major Pharma Collaboration Milestone Payments 战略合作已成为AI制药公司的重要收入来源，减少了直接融资需求：Strategic collaborations have become crucially paramount income origins for AI pharma firms dynamically, lowering primary direct financing requisites comprehensively: 合作方 / Partner 受益公司 / Benefiting    Co 合作内容 /    Collaboration Content 已收/合同金额 /    Revenue/Contract Value Roche/Genentech Recursion Phenomap数据合作 / Phenomap   data pipeline collab 2.13亿美元（已收） / $213M   (Received) Sanofi Recursion 小分子药物发现项目 / Small   molecule drug discovery initiatives 1.34亿美元（已收） / $134M   (Received) Merck KGaA BenevolentAI 药物发现合作 /   Foundational drug discovery collaborations 5.94亿美元（合同总额） / $594M   (Total contract sum) Menarini Insilico Medicine 候选药物授权（MEN2501+MEN2312）/ Candidate   licenses 合计>800万美元里程碑（已收）/ >$8M   combined milestones (Received) Eli Lilly Isomorphic Labs 药物发现合作 /   Foundational drug discovery collabs 未披露 / Undisclosed Novartis Isomorphic Labs 药物发现合作 /   Foundational drug discovery collabs 未披露 / Undisclosed 惠氏/辉瑞等 / Pfizer, etc. 晶泰科技 / XtalPi 固态形式预测服务 / Solid   format tracking schemas 服务费模式 / Service fee   frameworks 图12：主要战略合作里程碑付款对比Figure 12: Comparison of Major Strategic Collaboration Milestone Payments6.4 并购整合：Recursion收购Exscientia6.4 M&A Integration: Recursion acquires Exscientia 属性 / Attribute 详情 / Details 宣布时间 / Announced 2024年 / 2024 完成时间 / Completed 2024年11月 / Nov 2024 交易价值 / Deal Value 约6.88亿美元（全股票交易） / Approx. $688   million (All-stock transaction) 战略逻辑 / Strategic   Rationale Recursion获Exscientia的AI药物化学平台（Centaur Chemist）+ 临床管线 / Recursion   absorbs Exscientia’s distinct AI medicinal chemistry platform parameter   matrix dynamically (Centaur Chemist) + pipelines. 整合结果 / Integration   Results 2024-2025年终止3个临床项目（REC-2282、REC-994、REC-3964），优化管线至5个核心项目 / Terminated   3 distinct pipelines effectively mapping between 2024-2025; pipeline   structurally narrowed to 5 pure core projects. 行业信号：此次并购标志着AI制药行业进入整合期。资金压力下，中小型AI制药公司面临被大平台收购或与大型药企被动合作的压力。Industry Signal: This highly significant M&A explicitly flags that the AI pharma arena functionally enters its core consolidation stage systematically. Tracking escalating financial pressures routinely, small-to-midsized AI drug discovery organizations fundamentally face aggressive mandates for overarching acquisitions directly via broader platforms, or entirely passive alliances with massive legacy pharmaceutical organizations outright.6.5 主要投资机构6.5 Key Investment Institutions 类型 / Type 代表机构 / Representative Institution 代表投资 / Representative Investments 顶级VC / Top VC Thrive   Capital Isomorphic   Labs A轮领投 / Isomorphic Labs Series A Lead ARCH   Venture Partners Xaira Therapeutics联合创始 / Xaira Therapeutics Co-Founder Flagship   Pioneering Generate Biomedicines孵化 / Generate Biomedicines Incubator a16z (bio) 多家AI生物科技 / Multiple AI biotechs 战略投资 / Strat Investments Nvidia Recursion（5,000万美元，2023年）/ Recursion ($50M, 2023) Alphabet Isomorphic   Labs Eli Lilly Insilico Medicine（基石）/ Insilico Medicine (Cornerstone) 腾讯 / Tencent Insilico Medicine（基石）/ Insilico Medicine (Cornerstone) 淡马锡 / Temasek Insilico Medicine（基石）/ Insilico Medicine (Cornerstone) -–第七章：竞争格局与区域分布Chapter 7: Competitive Landscape and Regional Distribution7.1 全球三极格局7.1 The Tri-Polar Global Layout 美国：最大单一市场（约56%份额），资本最集中，监管环境相对开放。代表公司：Recursion（RXRX）、Relay Therapeutics（RLAY）、Schrödinger（SDGR）、Xaira、Atomwise。大型药企（辉瑞、礼来、BMS）AI内部能力建设最强。United States: Largest singular core market (~56% structural share), distinctly harboring optimal capital density mappings precisely framing open regulatory perimeters dynamically. Representative entities continuously explicitly function: Recursion (RXRX), Relay Therapeutics (RLAY), Schrödinger (SDGR), Xaira, Atomwise. Massive pharmaceutical organizations directly trace highly formidable internalized AI capacities explicitly currently. 中国/香港：增速最快（CAGR 21.1%），政策支持力度大，临床资源丰富（患者入组速度快），劳动力成本低。代表公司：英矽智能（3696.HK）、晶泰科技（2228.HK）、望石智慧、分子之心等。NMPA监管灵活性提高，Rentosertib中国Phase 2a数据令全球关注中国作为AI药物试验基地的价值。China/Hong Kong: The fundamentally fastest growing framework completely mapping explicitly globally entirely (CAGR 21.1%); distinctly tracking highly aggressive policy supports continuously. Massive baseline variables scale functionally optimized enrollment rates directly, tracing sharply reduced labor cost matrices structurally. Key bodies systematically frame functionality comprehensively: Insilico Medicine (3696.HK), XtalPi (2228.HK), StoneWise, MoleculeMind. Flexible regulatory parameters distinctly empower operations continually tracking NMPA adjustments, bringing overwhelming global significance recognizing Chinese footprints validating purely fundamental trial-base scopes tracking specifically Rentosertib metrics directly. 欧洲：学术实力强（牛津、剑桥等），监管框架相对严格但在积极演化。代表：Isomorphic Labs（伦敦）、BenevolentAI（伦敦，但已大幅收缩）。欧洲AI制药公司规模整体小于美国，主要以大型药企（诺华、罗氏、阿斯利康）为依托。Europe: Retains formidable fundamental academic power dynamics totally (Oxford, Cambridge). Regulatory outlines remain robustly rigid albeit incrementally tracking adaptive configurations directly. Representative groups routinely cover systematically Isomorphic Labs (London), BenevolentAI (London, deeply regressed operationally). European AI pharmaceutical entity scopes trace substantially narrower metric bandwidths contrasted aggressively against America, relying robustly specifically upon vast pharma backings structurally (Novartis, Roche, AstraZeneca).7.2 竞争维度分析7.2 Competitive Dimensionality Matrices 维度 / Dimension AI原生公司 / AI-Native Cos 大型药企内部AI / Big    Pharma Internal AI 传统CRO转型 / Traditional    CRO Pivots 临床管线深度 / Clinical pipeline depth 薄（1-5个临床项目）/ Thin (1-5 proj) 厚（数十个管线）/ Thick (Dozens) 无自有管线 / No native pipeline 平台技术深度 / Platform tech depth 深（专注AI优先）/ Deep (AI-first) 正在快速追赶 / Catching up rapidly 浅（外包转型）/ Shallow (Outsource pivots) 资金实力 / Funding capacity 有限（需持续融资）/ Limited (Needs VC) 雄厚 / Massive 有限 / Limited 数据优势 / Baseline data vectors 公开数据为主 / Predominantly public 大量专有临床数据 / Private massive clinical sets 实验数据 / Exclusively experimental data 速度 / Velocity 快（灵活）/ Fast (Flexible) 慢（官僚体制）/ Slow (Bureaucracy) 中等 / Moderate 风险承担能力 / Risk tolerance levels 高（初创文化）/ High (Startup culture) 低（关注ROI）/ Low (Strict ROI mapping) 中等 / Moderate 关键竞争趋势：Crucial Trend Metrics:1. 大型药企AI内化加速：辉瑞、礼来、诺华均在大力建设内部AI能力，威胁外部AI平台的市场空间 1.Accelerated mass pharma AI internalization: Pfizer, Novartis uniquely driving tremendous internal builds threatening explicit vendor pathways globally out-phasing platform boundaries fundamentally.2. 平台型公司面临"证明价值"压力：没有临床验证的平台型公司融资难度在增加（BenevolentAI教训） 2.Platform frameworks face aggressive “prove-worth” parameters: Zero clinical validity dramatically forces financing collapses fundamentally precisely tracking BenevolentAI metrics actively.3. 管线型公司更受投资者青睐：Insilico（有Rentosertib数据）比纯平台型公司估值溢价更高 3.Pipeline entities structurally vastly favored specifically by capital investors completely directly: Insilico (holding pristine tracking data exclusively structurally) yields monumentally expanded valuation premiums precisely overriding exclusively pure-platform entities dynamically entirely. 图13：全球AI制药竞争格局矩阵Figure 13: Global AI Drug Discovery Competitive Landscape Matrix -–第八章：风险与挑战Chapter 8: Risks and Challenges8.1 临床转化率：AI的真正考验尚未到来8.1 Clinical Translation Rate: The Ultimate AI Test Has Not Yet Arrived AI在临床前阶段的加速优势已有一定验证（更少化合物、更短时间），但临床成功率仍是最大未知数：AI systematically manifests fundamentally validated acceleration parameters tracking precisely explicitly in sweeping preclinical variables distinctively (reduced structural compounds precisely mapped inside minimized timelines entirely), yet ultimate clinical success metrics remain dramatically unknown predominantly:- DSP-1181（Exscientia）：全球首个AI临床药物，Phase 1后停止研发- DSP-1181 (Exscientia): The world’s first AI drug abruptly halted exactlyupon wrapping Phase 1 outright.- REC-994（Recursion）：Phase 2有积极数据，但已去优先级- REC-994 (Recursion): Generating dynamically positive Phase 2 outputs precisely but explicitly purely deprioritized heavily exactly.- BenevolentAI临床资产：多个失败，导致估值崩塌- BenevolentAI purely clinical asset branches: Massive multiple failures extensively directly triggered complete structural valuation collapses outright precisely.- Rentosertib：Phase 2a有积极信号，但Phase 2b/3（更大样本、更严格终点）才是真正的考验- Rentosertib: Demonstrates positively profound Phase 2a signal sweeps precisely, albeit ultimate tests distinctly map essentially against massive strict Phase 2b/3 requirements precisely dynamically fundamentally. AI加速了候选发现，但不改变临床试验本身的生物学挑战（靶点验证、患者异质性、ADMET复杂性）。目前Phase 1成功率提升（80-90% vs. 40-65%）的数据可能存在选择偏倚，需更多数据验证。AI explicitly accelerates purely discovery nodes heavily, completely absolutely preserving distinctly ultimate fundamental rigidities inherently tracking pure fundamental baseline biology complexities absolutely dynamically wholly definitively exclusively directly routinely strictly entirely explicitly broadly significantly exactly absolutely routinely fully natively thoroughly largely definitively highly critically largely largely functionally natively explicitly globally entirely (targeting validations broadly, heterogeneity variables routinely structurally heavily thoroughly deeply actively, and vastly extensive toxicity ADMET intricacies universally broadly overwhelmingly completely fully natively exactly deeply integrally largely). Pre-existing Phase 1 success rate shifts fundamentally structurally (80-90% mapped vs purely legacy 40-65% bands fully actively deeply) intrinsically reflect deeply biased selection data vectors routinely fundamentally demanding explicit mass verifications structurally inherently completely universally extensively functionally broadly entirely thoroughly strictly distinctly completely.8.2 数据质量与可及性8.2 Data Quality and Accessibility - 训练数据主要来自公开数据库，偏向已知靶点和化学骨架- Training data pools fundamentally predominantly map from open domains purely structurally exclusively routinely biasing extensively essentially explicitly against established target norms exactly strictly tracing historical scaffolds entirely natively completely natively highly inherently fully.- 高质量专有数据（大型药企的历史失败数据）不共享，限制AI模型泛化能力- Optimal quality structural private data broadly uniquely mapping historical failures explicitly globally inside colossal legacy pharmaceutical vaults remain broadly unshared fully, thoroughly stunting generic holistic AI extrapolation limits essentially absolutely completely definitively thoroughly routinely explicitly precisely entirely fundamentally broadly deeply.- 中国、印度等国家的临床数据质量和标准化程度不一- Disparate systemic qualities inherently trace universally fully explicitly broadly directly spanning vast clinical data matrices generated exclusively totally mapping thoroughly completely natively wholly purely distinctly directly fully completely precisely natively explicitly universally globally tracing completely thoroughly extensively natively holistically deeply fully structurally.8.3 AI可解释性挑战8.3 The AI Interpretability Challenge - 监管机构要求可解释性：为什么AI选择这个分子？为什么AI预测这个靶点有效？- 深度学习模型的"黑箱"特性与FDA对机制透明度的要求存在张力- 可解释AI（XAI）在制药领域的发展仍处于早期- 8.4估值泡沫风险8.4 Valuation Bubble Risks 风险因素 / Risk Factor 说明 / Description 商业化遥远 / Remote Commercialization 多数公司距产品化收入仍5年+ / Majority 5+ years away entirely 临床失败惩罚 / Clinical Failure Punishments 单个资产失败可导致估值暴跌90%+ / Single failures   crush val 90%+ 大药企内化威胁 / Deep Pharma Internalization 礼来/辉瑞等建立内部AI能力，压缩外包需求 / Pfizer/Lilly natively   crushing external structures 利率环境 / Interest Matrices 高利率下，长周期亏损公司估值承压 / High rates   universally suppress cash burners completely 竞争加剧 / Hyper Competition Xaira、Isomorphic进入，供给侧竞争激烈 / Xaira/Isomorphic   fundamentally sparking supply shocks totally8.5 监管不确定性8.5 Regulatory Uncertainties - FDA最终AI指南尚未发布（预计2026年Q2）- 国际监管框架不统一（ICH正在制定，预计2027年）- 中国NMPA对AI设计药物的加速审评路径细则仍待明确- 监管更严格化的风险（若出现重大安全事件）-–第九章：展望2026-2028年Chapter 9: Outlook 2026-20289.1 关键催化剂时间表9.1 Key Catalysts Timeline 时间节点 / Timeline 事件 / Event 影响等级 / Impact    Level 2026年Q1-Q2 / 2026 Q1-Q2 FDA最终AI监管指南发布 / FDA final AI regulatory   guidelines released 政策级 ★★★★☆ / Policy Level 2026年Q2 / 2026 Q2 REC-4881 FAP向FDA申请注册路径磋商 / REC-4881 FAP applies for FDA   registry path consultation 公司级 ★★★★☆ / Company Level 2026年Q2-Q3 / 2026 Q2-Q3 Isomorphic Labs首批临床数据公布（预期）/ Isomorphic Labs initial   clinical data release (Expected) 行业级 ★★★★★ / Industry Level 2026年H1 / 2026 H1 Rentosertib   Phase 2b/3启动（全球多中心）/ Rentosertib Phase 2b/3 initiation (Global multicenter) 行业级 ★★★★★ / Industry Level 2026年H2 / 2026 H2 RLY-2608 Phase III正式启动 / RLY-2608 Phase III official   initiation 行业级 ★★★★★ / Industry Level 2026年H2 / 2026 H2 REC-7735（PI3Kα）IND申请 / REC-7735 (PI3Kα) IND application 公司级 ★★★☆☆ / Company Level 2027-2028年 / 2027-2028 Rentosertib   Phase 2b顶线数据（预期）/ Rentosertib Phase 2b topline data (Expected) 行业级 ★★★★★ / Industry Level 2027-2028年 / 2027-2028 RLY-2608 Phase III中期分析 / RLY-2608 Phase III interim   analysis 行业级 ★★★★★ / Industry Level 2027-2028年 / 2027-2028 首个AI设计药物FDA审评申请（预期）/ First AI-designed drug FDA review application (Expected) 历史级 ★★★★★ / Historic Level9.2 技术演进预测9.2 Technology Evolution Predictions 蛋白质设计进化（2026-2028年）：Evolution of Protein Design (2026-2028):- "AlphaFold 4"商业化：动态构象预测能力提升，使变构位点药物设计成熟- “AlphaFold 4” commercialization: Enhanced dynamic conformational prediction capabilities mature allosteric site drug designs.- 蛋白质从头设计（Xaira/Baker）：首批临床候选预期2027-2028年进入Phase 1- De novo protein design (Xaira/Baker): Initial clinical candidates expected to enter Phase 1 in 2027-2028.- 蛋白质-蛋白质相互作用（PPI）靶点：传统化学无法成药，AI蛋白质设计将开辟新赛道- PPI targets: Traditionally undruggable, AI protein design will open this new frontier. 生成AI深化（2026-2028年）：Deepening of Generative AI (2026-2028):- 多模态分子生成（蛋白质+小分子+核酸）成为主流- Multimodal molecule generation (protein + small molecule + nucleic acid) becomes mainstream.- “AI科学家”（自主实验规划+机器人执行闭环）早期商业化- “AI Scientist” (autonomous experimental planning + robot execution loop) early commercialization.- 基于LLM的靶点发现成为大型药企标配- LLM-based target discovery becomes standard for large pharmaceutical companies. 临床预测AI（2026-2028年）：Clinical Prediction AI (2026-2028):- AI患者分层在临床试验设计中广泛应用- AI patient stratification widely applied in clinical trial design.- AI临床终点预测（通过生物标志物）开始进入监管申请- AI clinical endpoint prediction (via biomarkers) begins explicitly entering regulatory filings.- 数字孪生（Digital Twin）在药物安全性评估中试点- Digital Twin piloted in drug safety assessments.9.3 首个AI设计药物获批预测9.3 Predictions on the First Approved AI-Designed Drug 最快路径预测：Fastest Path Predictions:1. Rentosertib（Insilico，IPF）：若Phase 2b/3成功，2028-2030年具备申请NDA的可能。IPF是孤儿病，可享快速审评通道。中国NMPA路径或更快。 1.Rentosertib (Insilico, IPF): If Phase 2b/3 succeeds, an NDA application is possible between 2028-2030. IPF is an orphan disease eligible for fast-track review. The NMPA path in China might be even faster.2. RLY-2608（Relay，乳腺癌）：Phase III 2026年启动，若2027-2028年中期数据积极，2029-2030年可能提交申请。 2.RLY-2608 (Relay, Breast Cancer): Phase III initiates in 2026; if interim data in 2027-2028 is positive, submissions could occur in 2029-2030.3. Zasocentinib（Schrödinger/Takeda，IBD）：Phase III已在进行中，2027年可能是申请时间窗口。 3.Zasocentinib (Schrödinger/Takeda, IBD): Phase III is currently ongoing; 2027 could definitively be the application time window. 市场主流预测：首个明确标注"AI设计"的药物获FDA批准，大概率发生在2028-2030年。Mainstream Market Forecast: The first drug explicitly labeled as “AI-designed” is highly likely to be approved by the FDA between 2028-2030. -–附录一：完整临床管线汇总Appendix I: Complete Clinical Pipeline Summary 药物名称 / Drug 开发公司 / Co. 靶点 / Target 适应症 / Indication 阶段 / Phase AI贡献 / AI Contribution 状态 / Status Zasocentinib (TAK-279) Schrödinger/Takeda TYK2 溃疡性结肠炎/克罗恩病 / UC/CD Phase III 分子设计优化 / Molecule   design opt. 活跃 / Active RLY-2608 Relay Therapeutics PI3Kα 转移性乳腺癌 / Metastatic   breast cancer Phase III计划 / Planned 构象选择性设计 /   Conformation-selective design 活跃 / Active Rentosertib (ISM001-055) Insilico Medicine TNIK 特发性肺纤维化 / IPF Phase 2a✓ 靶点+分子全AI发现 / Full AI   target+molecule 规划Phase 2b/3 /   Planned Phase 2b/3 REC-4881 Recursion MEK1/2 家族性腺瘤性息肉病 / FAP Phase 2 Phenomics靶点识别 / Phenomics   target ID 活跃 / Active REC-994 Recursion — 脑海绵状血管畸形 / CCM Phase 2 表型AI / Phenotypic   AI 去优先级 /   Deprioritized RLY-1971 Relay Therapeutics SHP2 实体瘤 / Solid tumors Phase II 构象设计 /   Conformational design 活跃 / Active RLY-5836 Relay Therapeutics FGFR2 胆管癌 /   Cholangiocarcinoma Phase II 选择性设计 / Selectivity   design 活跃 / Active EXS-4318 Recursion (前Exscientia) CDK7 肿瘤 / Oncology Phase II AI药物化学 / AI medicinal   chemistry 活跃 / Active BEN-8744 BenevolentAI PDE10A 溃疡性结肠炎 / UC Phase I/II 知识图谱靶点发现 / KG   target discovery 数据积极 / Data   positive MEN2501 Insilico→Menarini 未披露 / Undisclosed 实体瘤 / Solid tumors Phase I✓ 全AI设计 / Full AI design 2026.02里程碑 / Feb 2026 MEN2312 Insilico→Menarini KAT6 乳腺癌 / Breast cancer Phase I 全AI设计 / Full AI design 活跃 / Active ISM5411 Insilico Medicine 未披露 / Undisclosed 系统性红斑狼疮 / SLE Phase 1/2 全AI设计 / Full AI design 活跃 / Active REC-617 Recursion CDK7 铂耐药卵巢癌 / Ovarian   cancer Phase 1/2 11个月发现 / 11-month   discovery 活跃 / Active REC-1245 Recursion RBM39降解剂 / RBM39   degrader 肿瘤 / Oncology Phase 1 AI驱动 / AI-driven 活跃 / Active REC-3565 Recursion MALT1 — Phase 1 AI驱动 / AI-driven 活跃 / Active REC-4539 Recursion LSD1 — Phase 1 AI驱动 / AI-driven 活跃 / Active ISM3312 Insilico Medicine 3CLpro COVID-19 Phase I AI设计蛋白酶抑制剂 /   AI-designed protease inhibitor 活跃 / Active EXS21546 Recursion (前Exscientia) A2a受体 / A2a receptor 肿瘤免疫 / IO Phase I (2020年启 / 2020) AI药物设计 / AI drug   design 活跃 / Active DSP-0038 Recursion (前Exscientia) 5-HT1a/2a 阿尔茨海默症精神病 / AD   Psychosis Phase I 多靶点AI设计 / Multi-target   AI design 活跃 / Active SGR-1505 Schrödinger MALT1 血液肿瘤 / Hematologic   tumors Phase I 物理+ML设计 / Physics+ML   design 活跃 / Active SGR-2921 Schrödinger CDC7 实体瘤 / Solid tumors Phase I 物理+ML设计 / Physics+ML   design 活跃 / Active DSP-1181 Exscientia (历史 / History) 5-HT1A 强迫症 (OCD) Phase I (2020年) 首个AI临床药 / First AI   clinical drug 已停止 / Halted -–附录二：公司财务数据汇总Appendix II: Corporate Financial Data Summary 公司 / Company 股票代码 / Ticker 2025年营收 / 2025 Rev 净亏损/盈利 / Net    Loss/Profit 现金储备 / Cash Reserves Cash Runway Insilico Medicine 3696.HK ~8,583万美元(2024年) ~$85.83M (2024) -1,700万美元(2024年) -$17M (2024) ~3亿美元+ ~$300M+ 2027年+ / 2027+ Recursion RXRX 7,470万美元 $74.7M -6.448亿美元 -$644.8M 7.539亿美元 $753.9M 2028年初 / Early 2028 BenevolentAI BAI — — ~约8,500万英镑市值 ~£85M Market   Cap 有限 / Limited Schrödinger SDGR ~2亿美元(软件) ~$200M   (Software) 净亏损 Net Loss ~7亿美元+ ~$700M+ 2027年+ / 2027+ Relay Therapeutics RLAY — 净亏损 Net Loss ~约10亿美元 ~$1B 2027年+ / 2027+ -–附录三：数据来源与可信度评级Appendix III: Data Sources and Credibility Ratings 来源 / Source 类型 / Type 可信度 / Credibility 主要用途 / Main Use Nature Medicine (2025, Vol.31, pp.2602-2610) 同行评审学术期刊 /   Peer-reviewed Academic Journal ⭐⭐⭐⭐⭐ Rentosertib Phase 2a数据 / Rentosertib   Phase 2a data Insilico Medicine香港上市招股书 (2025年12月) / HK Listing   Prospectus 官方监管文件 / Official   Regulatory Filings ⭐⭐⭐⭐⭐ 公司财务/管线数据 /   Financials/Pipeline data Recursion 2025年全年财报公告 (2026年2月25日) / 2025 Annual   Earnings 官方IR公告 / Official IR   Announcements ⭐⭐⭐⭐⭐ Recursion财务数据 / Recursion   financials Axis Intelligence 2026年AI临床管线分析 / 2026   Pipeline Analysis 专业行业研究机构 /   Professional Industry Research ⭐⭐⭐⭐☆ 173个项目统计数据 /   Statistics on 173 projects ClinicalTrials.gov 美国FDA官方数据库 / Official US   FDA Database ⭐⭐⭐⭐⭐ 临床试验注册信息 / Clinical   trial registry info Precedence Research / Mordor Intelligence等 / etc. 市场研究机构 / Market   Research Firms ⭐⭐⭐☆☆ 市场规模预测 / Market   size forecasts Fortune Business Insights 商业媒体/研究机构 / Business   Media/Research ⭐⭐⭐☆☆ 行业趋势 / Industry   trends FierceBiotech / BioBriefs 行业专业媒体 / Industry   Media ⭐⭐⭐⭐☆ 公司动态 / Company   updates 公司官网 / IR官网 / Company   websites / IR sites 一手来源 / Primary   Source ⭐⭐⭐⭐⭐ 公司官方信息 / Official   company info 新华财经 / 香港交易所公告 / Xinhua   Finance / HKEX filings 官方交易所文件 / Official   Exchange Filings ⭐⭐⭐⭐⭐ 上市公司信息 / Listed   company info -– *本报告编制完成时间：2026年3月1日*Report compilation completion date: March 1, 2026 *数据截止：2026年2月28日*Data cut-off: February 28, 2026 *报告版权所有，转载请注明来源*Copyright of the report belongs to ZHOUHAO; please indicate the source when reprinting. 若要获取报告全文docx及pdf文档，请关注公众号，后台免费领取。

扫码查看“数智医药”专题当传统医药行业还陷在“十年研发周期、十亿美元投入”的“双十困局”时，人工智能（AI）正以破局之势撕开突破口，重新构建行业的效率边界及能力边界。和铂医药AI战略再升级，前不久正式发布其AI平台Hu-mAtrIxTM驱动的首个全人源AI HCAb模型——全人源重链抗体（HCAb）生成和筛选模型。同时，和铂医药还发起成立了AI+生物医药生态圈联盟（以下简称联盟），首批成员包括孚腾资本、英矽智能、分子之心、阿惟络医疗科技等。从制药环节的单点突破，到联盟式的“全链条”创新，AI将为医药行业打开全新的想象空间。AI不仅降本增效，还能“变不可能为可能”AI、大数据等新技术正以前所未有的速度重塑药物研发模式，驱动医药行业变革。“抗体药物研发中，AI可以辅助优化抗体的亲和力、特异性和稳定性，从而提高药物疗效并降低副作用。另外，AI技术还可以帮助探索新的药物研发领域和靶点。一些传统方法难以攻克的疾病，如神经退行性疾病、罕见病等，通过AI分析和预测，可以发现新的治疗策略和药物分子，为这些疾病的治疗带来新希望。”和铂医药创始人、董事长兼首席执行官王劲松说。当前，AI已应用于靶点发现、分子设计、临床试验等药物研发的多个环节。分子之心创始人兼首席执行官许锦波表示，在分子设计阶段，AI不仅可以降本增效，还能“变不可能为可能”。“比如一些功能蛋白设计，如果用传统计算工具，要做很长时间，也有可能做不出来。AI让高度复杂且非自然界存在的蛋白设计成为可能。”英矽智能一款完全利用AI发现和设计的药物Rentosertib已经处于Ⅱ期临床试验阶段。该公司联席首席执行官兼首席科学官任峰介绍，目前AI已经应用到英矽智能药物研发的多流程中，包括靶标发现、小分子设计，以及临床试验方案优化等。任峰以靶标发现为例：“感触最深的是，AI可以帮我们去做一些原创性很强的东西。过去找靶点，往往是通过已经发表的文献，找蛋白或者基因功能，然后再试图把功能和疾病联系起来。AI的优势在于能够处理海量的患者多组学数据，并将其与健康人的数据进行差异对比，直接定位致病因素，找到携带突变的基因。这种‘逆向’方式，极大提高了研发效率。”临床试验是药物研发流程中最耗时、也最耗资金的环节。Evinova中国总经理张杰表示，AI在赋能临床开发平台方面，已经展现出巨大潜力。“AI在做患者画像（Patient Profile）与临床方案（Protocol）匹配时，准确率已经超过80%，接近人类专家水平，这意味着AI能高效筛选出最合适的潜在患者。在加速患者招募方面，AI的介入可以节省约40%时间。”此外，他表示，AI agent（人工智能体）在临床方案设计方面也已经开展了很多探索。“绿皮火车”变“高铁”，不能只换轨道和信号灯早在此次联盟成立之前，和铂医药已经开始与相关企业就AI技术开展合作。例如，今年2月，和铂医药与英矽智能达成战略合作，共同推进AI赋能的抗体发现算法和应用开发。和铂医药具有行业领先的技术平台、专有数据集以及丰富抗体开发经验，英矽智能则在构建一体化AI驱动药物研发平台方面拥有技术优势。谈及为何要成立AI联盟，王劲松作了一个比喻，“从绿皮火车变成高铁，不是只换轨道和信号灯就可以。我们看到，多数公司都是在制药链条的某个环节深挖AI应用，但这远不能发挥AI的全部价值。”在王劲松看来，AI待挖掘的最大潜在价值，或许在于解决新药研发中最具挑战性的“转化预判”问题。传统的药物研发失败率极高，细胞模型和动物模型等临床前数据与人体临床数据之间往往存在巨大断层，导致前期判断的有效性和安全性在人体内经常失效，这也是药物研发的痛点和风险所在。以动物模型为例，其与人类疾病的发生、发展，以及药物反应等方面存在本质差异，尤其是动物的肿瘤微环境无法真实模拟人体情况。业内人士表示，AI有望大大提升新药研发中“转化预判”的准确率，在更早期阶段预测分子的临床表现，让行业真正告别“双十困局”。当然，前提是有足够的数据，而且是能够覆盖研发全链条的数据。这也是联盟成立的原因所在——让更多上下游数据互通。可以看到，联盟中既有制药企业，也有医疗机构、科研院所。“大型制药公司、创新药企业、科研院所，手中积累的数据都是不一样的。制药公司更多的是临床数据，创新药企业可能是一些实验数据，院校或者科研机构更多的是研发数据。如何推动多机构之间的数据互通，是未来要聚焦的一个问题。这些数据对于每家机构来说都是关键资产，未来或许可以通过隐私计算等技术打破壁垒。”亚马逊云科技中国医疗及生命科学行业解决方案团队负责人姚亮表示。（文据新华财经，配图为AI生成）

当传统医药行业还陷在“十年研发周期、十亿美元投入”的“双十困局”时，人工智能（AI）正以破局之势撕开突破口，重新构建行业的效率边界及能力边界。和铂医药AI战略再升级，前不久正式发布其AI平台Hu-mAtrIxTM驱动的首个全人源AI HCAb模型——全人源重链抗体（HCAb）生成和筛选模型。同时，和铂医药还发起成立了AI+生物医药生态圈联盟（以下简称联盟），首批成员包括孚腾资本、英矽智能、分子之心、阿惟络医疗科技等。从制药环节的单点突破，到联盟式的“全链条”创新，AI将为医药行业打开全新的想象空间。AI不仅降本增效，还能“变不可能为可能”AI、大数据等新技术正以前所未有的速度重塑药物研发模式，驱动医药行业变革。“抗体药物研发中，AI可以辅助优化抗体的亲和力、特异性和稳定性，从而提高药物疗效并降低副作用。另外，AI技术还可以帮助探索新的药物研发领域和靶点。一些传统方法难以攻克的疾病，如神经退行性疾病、罕见病等，通过AI分析和预测，可以发现新的治疗策略和药物分子，为这些疾病的治疗带来新希望。”和铂医药创始人、董事长兼首席执行官王劲松说。当前，AI已应用于靶点发现、分子设计、临床试验等药物研发的多个环节。分子之心创始人兼首席执行官许锦波表示，在分子设计阶段，AI不仅可以降本增效，还能“变不可能为可能”。“比如一些功能蛋白设计，如果用传统计算工具，要做很长时间，也有可能做不出来。AI让高度复杂且非自然界存在的蛋白设计成为可能。”英矽智能一款完全利用AI发现和设计的药物Rentosertib已经处于Ⅱ期临床试验阶段。该公司联席首席执行官兼首席科学官任峰介绍，目前AI已经应用到英矽智能药物研发的多流程中，包括靶标发现、小分子设计，以及临床试验方案优化等。任峰以靶标发现为例：“感触最深的是，AI可以帮我们去做一些原创性很强的东西。过去找靶点，往往是通过已经发表的文献，找蛋白或者基因功能，然后再试图把功能和疾病联系起来。AI的优势在于能够处理海量的患者多组学数据，并将其与健康人的数据进行差异对比，直接定位致病因素，找到携带突变的基因。这种‘逆向’方式，极大提高了研发效率。”临床试验是药物研发流程中最耗时、也最耗资金的环节。Evinova中国总经理张杰表示，AI在赋能临床开发平台方面，已经展现出巨大潜力。“AI在做患者画像（Patient Profile）与临床方案（Protocol）匹配时，准确率已经超过80%，接近人类专家水平，这意味着AI能高效筛选出最合适的潜在患者。在加速患者招募方面，AI的介入可以节省约40%时间。”此外，他表示，AI agent（人工智能体）在临床方案设计方面也已经开展了很多探索。“绿皮火车”变“高铁”，不能只换轨道和信号灯早在此次联盟成立之前，和铂医药已经开始与相关企业就AI技术开展合作。例如，今年2月，和铂医药与英矽智能达成战略合作，共同推进AI赋能的抗体发现算法和应用开发。和铂医药具有行业领先的技术平台、专有数据集以及丰富抗体开发经验，英矽智能则在构建一体化AI驱动药物研发平台方面拥有技术优势。谈及为何要成立AI联盟，王劲松作了一个比喻，“从绿皮火车变成高铁，不是只换轨道和信号灯就可以。我们看到，多数公司都是在制药链条的某个环节深挖AI应用，但这远不能发挥AI的全部价值。”在王劲松看来，AI待挖掘的最大潜在价值，或许在于解决新药研发中最具挑战性的“转化预判”问题。传统的药物研发失败率极高，细胞模型和动物模型等临床前数据与人体临床数据之间往往存在巨大断层，导致前期判断的有效性和安全性在人体内经常失效，这也是药物研发的痛点和风险所在。以动物模型为例，其与人类疾病的发生、发展，以及药物反应等方面存在本质差异，尤其是动物的肿瘤微环境无法真实模拟人体情况。业内人士表示，AI有望大大提升新药研发中“转化预判”的准确率，在更早期阶段预测分子的临床表现，让行业真正告别“双十困局”。当然，前提是有足够的数据，而且是能够覆盖研发全链条的数据。这也是联盟成立的原因所在——让更多上下游数据互通。可以看到，联盟中既有制药企业，也有医疗机构、科研院所。“大型制药公司、创新药企业、科研院所，手中积累的数据都是不一样的。制药公司更多的是临床数据，创新药企业可能是一些实验数据，院校或者科研机构更多的是研发数据。如何推动多机构之间的数据互通，是未来要聚焦的一个问题。这些数据对于每家机构来说都是关键资产，未来或许可以通过隐私计算等技术打破壁垒。”亚马逊云科技中国医疗及生命科学行业解决方案团队负责人姚亮表示。（文据新华财经） 《中国医药报》社版权所有，未经许可不得转载使用。 (责任编辑：刘思慧)