作者:傅翔 薛超然
美工:何国红 罗真真
排版:马超
▶ 引言
在单域抗体药物开发中,等电点(pI)是决定抗体成药性的关键理化参数。然而,传统单域抗体常因等电点偏低而面临非特异性结合、聚集倾向高、体内稳定性差等成药性瓶颈,严重制约了抗体药物的开发效率与临床转化。三优生物依托十年技术积淀与海量实验数据积累,融合自主知识产权的AI算法与自动化筛选技术,成功推出AI-STAL 2.0高等电点单域抗体智筛系统。该平台通过“海量数据积累→AI模型持续训练调优→AI指导文库智能设计及自动化构建→自动化高通量筛选→AI驱动的成药性评估→干湿闭环迭代验证”的全链路智能体系。它将文库中单域抗体的等电点精准锁定在8.0-9.0,系统性排除翻译后修饰(PTM)位点,并实现高达98%的骨架人源化,真正从源头实现了成药性的全面优化。
通过将AI-STAL 2.0平台与三优智能新药加速器(SAI-DA)深度整合,研发效率被推向极致:全程自动化筛选,2周内即可完成多轮淘选与初筛,高效获得大量候选分子。针对不同难度靶点进行千亿级文库筛选,平均可获得35个序列独特、特异性强、成药性高的先导抗体。这一创新模式以“AI全流程赋能、自动化高效运行、智能化成药性评估”三位一体的智筛范式,为全球合作伙伴提供了高等电点、高稳定性、高人源化程度的优质先导分子,重新定义了单域抗体发现的未来。
▲ Fig.1 Full Intelligent High pI Single Domain Antibody Discovery Platform
一、文库背景
据Fortune Business Insights数据显示,单域抗体市场正迎来爆发式增长。2025年全球市场规模为4.067亿美元,预计2034年将增长至18.29亿美元,年复合增长率(CAGR)高达19.53%。同时,单域抗体文库服务市场增速达14.3%,远超传统抗体研发服务。凭借分子量小(约15 kDa)、组织穿透性强、结构稳定性高、生产成本低等独特优势,单域抗体已成为生物医药创新的前沿方向。
▲ Fig. 2 Global Nanobody Therapeutics Market Size, 2025-2034
然而,传统单域抗体在成药性上面临多重挑战:低等电点容易导致非特异性结合和聚集倾向,翻译后修饰位点影响抗体均一性与稳定性,动物源骨架可能引发免疫原性风险。研究表明,等电点(pI)在8.0-9.0区间的高等电点单域抗体,具有更强的抗聚集能力、更好的体内稳定性、更低的非特异性结合及更理想的纯化特性。然而,要构建同时兼具“高等电点”与“高功能多样性”的单域抗体库,仍需攻克多重技术瓶颈—包括等电点与功能活性的平衡、PTM位点的系统性排除、人源化程度的提升、文库多样性与成药性的兼顾,以及预测精度与实验验证之间差距。
为攻克这一系列瓶颈,三优生物凭借前瞻性的AI驱动技术布局与海量抗体数据积累,走出一条独特的技术路径:“海量数据积累→AI模型持续训练调优→AI指导文库智能设计及自动化构建→自动化高通量筛选→AI驱动的成药性评估→干湿闭环迭代验证”。在AI模型的精准指导下,文库设计实现了等电点的精准维持与PTM位点的系统性剔除。公司于2024年成功构建并验证了高等电点单域抗体库,并计划于2026年进一步扩容升级,推出超千亿高等电点单域抗体库,这一突破性进展标志着三优生物在单域抗体发现领域迈入了AI全流程赋能、数据驱动设计的全新阶段。
▲ Fig. 3 AI-Driven High pI Single Domain Antibody Discovery Workflow
二、文库组成
AI-STAL 2.0高等电点单域抗体库在万亿级人源化单域抗体库的基础上持续迭代创新,采用先进的人工智能算法对纳米抗体进行全序列优化与反向筛选。通过精准调控,文库中纳米抗体的等电点严格控制在8.0-9.0;同时系统性排除影响成药性的PTM位点。2026年,总库容可达2.00千亿。
▲ Fig. 4 Hundred‑Billion High‑pI Single‑Domain Antibody Library
三、文库特性
01
等电点精准锁定8.0-9.0
通过对千亿级高等电点单域抗体库筛选测序,序列分析结果表明,等电点(pI)在8.0-9.0范围内的分子占比显著提高。AI模型基于海量序列-pI关系数据,精准调控氨基酸组成,使文库分子的等电点严格锁定在目标区间。高等电点能够改善溶解度、降低聚集倾向并增强在体内的稳定性,从而显著提升候选药物的生物利用度与药代动力学特性。此外,该等电点范围内的分子通常表现出更低的非特异性结合,减少潜在的脱靶效应,进一步提升药物的安全性和有效性。
▲ Fig. 5 Comparison of pI 8.0–9.0 Molecules Distribution
02
系统性排除PTM位点
AI内置的PTM风险识别模块系统性地排除了影响成药性的翻译后修饰 (PTM) 位点,涵盖N-糖基化 (NxT/S) 、天冬酰胺脱酰胺 (NG/NS/QG/QS) 、天冬氨酸异构化 (DG/DS) 及水解位点 (DP) 等。序列分析结果显示,候选分子均不含PTM位点。无PTM位点意味着这些分子具有更高的结构稳定性与更低的免疫原性,显著提升了其成药性。同时,也降低了生产过程中出现异质性的风险,有助于简化纯化与质量控制流程。
▼ Table 1 Post-Translational Modifications
03
人源化程度高达98%
三优生物构建的千亿级高等电点单域抗体库具备高人源化特征。AI整合了130多个单域抗体分子的人源化项目实验数据,精准指导骨架区(FR区)的人源化设计。通过文库筛选,可直接获得人源化程度高达98%的单域抗体,无需进行额外的人源化改造流程,大幅缩短了项目周期。
▲ Fig. 6 Humanization Level Comparison
04
AI驱动的全面成药性评估
千亿级高等电点单域抗体库筛选获得的候选分子,需经过AI系统的全面成药性智能评估。AI模型基于海量历史项目数据,从多维度对候选分子进行系统分析与预测。可开发性预测:基于序列特征预测抗体表达量、溶解度、聚集倾向及热稳定性,提前识别潜在风险,大幅降低后期失败概率。免疫原性风险评估:通过分析序列中的T细胞表位、B细胞表位及MHC结合亲和力,预测并优先筛选低免疫原性分子。理化性质智能分析:自动预测等电点、疏水性、电荷异质性等关键理化参数,确保候选分子在纯化、制剂及储存等环节具有良好的可操作性。
经AI多维度智能评估后,优选的分子才进入真核表达验证阶段,开展全面的生化理化分析及细胞水平功能验证,从而获得真实有效的成药性数据。这一“AI初筛→湿实验精筛”的干湿接力模式,覆盖度广、筛选层次深,可高效过滤低质量分子,大幅提升进入实验验证阶段分子的“含金量”。
▼ Table 2 Developability of Antibodies
05
先导分子数量多
针对不同难度的靶点,该平台平均可获得35个序列独特、特异性强、成药性高的先导抗体,显著加速抗体药物的开发进程。该平台不仅适用于肿瘤、自身免疫性疾病等传统治疗领域,还可拓展至双特异性抗体、抗体偶联药物(ADC)、细胞治疗等前沿方向。
▲ Fig. 7 Number of Lead Molecules from Different Projects
四、代表案例
目的:在千亿级高等电点单域抗体库中筛选能够高亲和力结合BCMA(B细胞成熟抗原)的单域抗体分子。
背景:BCMA是肿瘤坏死因子受体超家族(TNFR)的重要成员,在恶性浆细胞疾病和自身免疫性疾病治疗中具有关键作用。BCMA靶向疗法已在临床取得显著进展,包括CAR-T细胞疗法(如Abecma, Carvykti)、抗体偶联药物(ADC)(如Blenrep)以及双特异性抗体(如Tecvayli)。药物作用机制均为特异性结合BCMA,从而介导肿瘤细胞的清除。
方法:利用AI-STAL 2.0智筛系统,从千亿级高等电点单域抗体库中经全程自动化筛选获得候选分子,随后通过AI进行全面的成药性智能评估后,优选分子进行真核表达。采用FACS分析候选抗体与huBCMA-HEK293细胞的亲和力;采用ELISA方法检测候选抗体是否阻断BAFF-NFC与huBCMA-ECD-His-Biotin蛋白的结合。
结果:如图所示,FACS分析结果显示,全部候选抗体的亲和力水平优于对照抗体。其中,候选抗体A001、A002、A003和A004能够阻断BAFF-NFC与huBCMA-ECD-His-Biotin蛋白的结合,阻断效果与对照抗体相当或优于对照抗体。
结论:经AI-STAL 2.0智筛系统筛选、评估与验证,成功获得细胞水平结合活性优异、阻断活性显著的高质量候选单域抗体分子。
▲ Fig. 8 Affinity testing of antibodies
▲ Fig. 9 Blocking testing of antibodies
五、总结展望
当前,AI驱动的抗体设计正加速从概念加速迈向产业化,前沿成果不断涌现。在这一浪潮中,三优生物的AI-STAL 2.0高等电点单域抗体智筛系统构筑起独具特色的技术壁垒,实现了AI智能化、筛选自动化、评估系统化三位一体的深度融合:AI贯穿建库、筛选、评估全流程,以海量数据持续训练调优模型,指导文库智能设计与候选分子精准筛选;SAI-DA全程自动化筛选,极短时间内完成多轮淘选与高通量初筛,高效覆盖海量候选分子;AI驱动的全面成药性评估,在湿实验之前对候选分子进行可开发性、免疫原性及理化性质等多维度智能分析与预测,确保进入真核验证阶段的分子“优中选优”。
该系统融合千亿级文库分子多样性、AI全流程赋能的智能发现引擎、全程自动化的高效筛选体系,以及高等电点(8.0-9.0)与无PTM位点带来的优异成药性,实现了98%超高人源化程度与平均35个优质先导分子的高效产出——这不仅是规模的突破,更是从“随机筛选”到“数据制导”、从“手工操作”到“自动运行”、从“经验判断”到“AI评估”的全方位方法论革命。
展望未来,三优生物将持续以海量实验数据喂养和优化AI模型,不断扩大抗体库资源与项目数据积累,让AI模型在更多真实数据的淬炼中持续进化;同时深化深度学习算法与自动化平台的整合,推动单域抗体从序列智能设计、自动化高效筛选到成药性精准预测的全链路升级。AI-STAL 2.0不只是一座取之不尽的分子宝库,更是一座覆盖度广、智能化高、自动化强的创新生物药源头智能引擎——能够精准定位靶点、高效产出候选分子、智能评估成药属性、全程护航药物开发,让每一次实验都更有价值、更有效率。通过这些创新举措,三优生物将显著降低高难度靶点单域抗体药物的开发难度,极大加速创新药物的研发进程,以前瞻性的技术布局和极致的研发效率赋能全球新药研发,为全球患者带来更高效、更安全的治疗选择。
AI-STAL 2.0 | "End-to-End Intelligent" High-pI Single-Domain Antibody Discovery System
▶ Introduction
In single-domain antibody (sdAb / Nanobody) drug discovery, the isoelectric point (pI) is a critical physicochemical parameter governing druggability and developability. However, conventional single-domain antibodies frequently face druggability bottlenecks—such as non-specific binding, high aggregation propensity, and poor in vivo stability—primarily due to low pI values. These limitations severely hinder therapeutic development and clinical translation. Leveraging a decade of technical expertise and vast experimental data, Sanyou Biopharmaceuticals has integrated proprietary AI algorithms with automated screening technologies to launch AI-STAL 2.0, an intelligent high-pI single-domain antibody discovery system. The platform operates on a closed-loop intelligent pipeline: Massive Data Accumulation → Continuous AI Model Training & Fine-Tuning → AI-Guided Intelligent Library Design & Automated Construction → High-Throughput Automated Screening → AI-Driven Developability Assessment → Dry-Wet Closed-Loop Iterative Validation. It precisely locks the pI of library antibodies within the range of 8.0–9.0, systematically eliminates post-translational modification (PTM) sites, and achieves up to 98% framework humanization, optimizing developability comprehensively right from the design phase.
By deeply integrating the AI-STAL 2.0 platform with Sanyou AI-Drug Accelerator (SAI-DA), R&D efficiency is driven to the extreme: fully automated screening completes multiple rounds of panning and primary screening within 2 weeks, rapidly yielding a large pool of candidate molecules. For targets across varying difficulty levels, screening the hundred-billion-scale library yields an average of 35 lead antibodies featuring unique sequences, high specificity, and superior developability. Embracing a paradigm of full-process AI empowerment, high-efficiency automated operation, and intelligent developability assessment, this innovative system provides global partners with high-pI, highly stable, and highly humanized lead candidates, redefining the future of single-domain antibody discovery.
▲ Fig.1 Full Intelligent High pI Single Domain Antibody Discovery Platform
I. Library Background
According to data from Fortune Business Insights, the global single-domain antibody market is experiencing explosive growth: valued at $406.7 million in 2025, it is projected to reach $1.829 billion by 2034, registering a compound annual growth rate (CAGR) of 19.53%. Meanwhile, the market for single-domain antibody library services is growing at 14.3%, significantly outstripping traditional antibody R&D services. Featuring a small molecular weight (~15 kDa), deep tissue penetration, high structural stability, and cost-effective production, single-domain antibodies have become a frontier in biopharmaceutical innovation.
▲ Fig. 2 Global Nanobody Therapeutics Market Size, 2025-2034
However, conventional single-domain antibodies encounter multiple developability challenges: Low pI values frequently cause non-specific binding and a high propensity for aggregation; PTM sites impair antibody homogeneity and stability; Animal-derived frameworks pose potential immunogenicity risks.
Studies demonstrate that high-pI single-domain antibodies with a pI between 8.0 and 9.0 exhibit stronger anti-aggregation capacity, enhanced in vivo stability, reduced non-specific binding, and ideal purification profiles. Nevertheless, constructing a single-domain antibody library that simultaneously achieves both a high pI and high functional diversity requires overcoming multiple technical hurdles: balancing pI with functional activity, systematically excluding PTM sites, elevating humanization levels, balancing diversity with developability, and bridging the gap between computational predictions and experimental validation.
To overcome these hurdles, Sanyou Bio has pursued a unique technological roadmap driven by forward-looking AI strategies and massive data accumulation: Massive Data Accumulation → Continuous AI Model Training & Fine-Tuning → AI-Guided Intelligent Library Design & Automated Construction → High-Throughput Automated Screening → AI-Driven Developability Assessment → Dry-Wet Closed-Loop Iterative Validation. Under the precise guidance of AI models, library design maintains targeted pI values while systematically removing PTM sites. Building on the successful construction and validation of its high-pI single-domain antibody library in 2024, Sanyou Bio plans to further expand and upgrade the platform in 2026 to launch an ultra-hundred-billion high-pI single-domain antibody library. This breakthrough marks Sanyou Bio's transition into a new era of data-driven, end-to-end AI-empowered antibody discovery.
▲ Fig. 3 AI-Driven High pI Single Domain Antibody Discovery Workflow
II. Library Composition
Iterated and innovated upon the trillion-scale humanized single-domain antibody library, the AI-STAL 2.0 High-pI Single-Domain Antibody Library utilizes advanced AI algorithms for full-sequence optimization and reverse screening of nanobodies. Through precise regulation, the pI of nanobodies in the library is strictly controlled within 8.0–9.0, while PTM sites detrimental to developability are systematically excluded. By 2026, the total library capacity will reach 200 billion (2.00 × 10¹¹).
▲ Fig. 4 Hundred‑Billion High‑pI Single‑Domain Antibody Library
III. Key Characteristics of the Library
01
Precise pI Locking at 8.0–9.0
Sequence analysis from high-throughput sequencing of screened candidates reveals a significantly increased proportion of molecules within the pI 8.0–9.0 range. Based on massive sequence–pI relationship datasets, the AI model precisely tunes the amino acid composition to lock the pI strictly within the target range. A high pI improves solubility, minimizes aggregation propensity, and enhances in vivo stability, thereby markedly elevating the bioavailability and pharmacokinetic (PK) properties of lead candidates. Furthermore, molecules within this pI range typically exhibit reduced non-specific binding, minimizing potential off-target effects and enhancing safety and efficacy.
▲ Fig. 5 Comparison of pI 8.0–9.0 Molecules Distribution
02
Systematic Exclusion of PTM Sites
The built-in AI PTM risk assessment module systematically eliminates post-translational modification (PTM) sites that compromise developability, including: N-glycosylation (NxT/S, where x ≠ P); Asparagine deamidation (NG/NS/QG/QS); Aspartic acid isomerization (DG/DS); Hydrolysis / cleavage sites (DP).
Sequence analyses confirm that the selected candidate molecules are completely free of PTM liabilities. The absence of PTM sites translates to higher structural stability, lower immunogenicity, and reduced risk of molecular heterogeneity during manufacturing, thereby streamlining purification and quality control (QC) workflows.
▼ Table 1. Excluded Post-Translational Modifications (PTMs)
03
Up to 98% Humanization Level
The hundred-billion high-pI single-domain antibody library constructed by Sanyou Bio features exceptionally high humanization. By integrating experimental data from over 130 nanobody humanization projects, the AI guides the precision design of framework regions (FRs). Screening directly yields single-domain antibodies with up to 98% humanization, eliminating the need for downstream humanization engineering and substantially shortening project timelines.
▲ Fig. 6 Humanization Level Comparison
04
AI-Driven Comprehensive Developability Assessment
Candidates derived from the hundred-billion high-pI single-domain antibody library undergo comprehensive computational developability assessment prior to wet-lab validation. Utilizing historical project datasets, the AI model systematically evaluates candidates across multiple dimensions: Developability Prediction: Predicts expression levels, solubility, aggregation propensity, and thermal stability based on sequence features, preemptively identifying risks and drastically reducing downstream attrition. Immunogenicity Risk Assessment: Analyzes T-cell epitopes, B-cell epitopes, and MHC binding affinity to prioritize candidates with low immunogenicity profiles. Physicochemical Property Profiling: Automatically calculates pI, hydrophobicity, charge heterogeneity, and other critical metrics to ensure favorable behavior during purification, formulation, and storage.
Only candidates that pass multi-dimensional AI evaluation advance to eukaryotic expression and functional testing in vitro (cell-based and biochemical assays). This relay model (AI Prescreening → Wet-Lab Refinement) provides extensive coverage and deep evaluation, efficiently filtering out inferior molecules and maximizing the success rate of experimental validation.
▼ Table 2. Experimental Developability Assessment of Antibodies
05
High Output of Lead Molecules
Across targets of varying technical difficulty, the platform yields an average of 35 unique, highly specific, and developable lead antibodies, significantly accelerating drug discovery programs. Beyond oncology and autoimmune diseases, the platform extends seamlessly to next-generation modalities, including bispecific antibodies, antibody-drug conjugates (ADCs), and cell therapies.
▲ Fig. 7 Number of Lead Molecules from Different Projects
IV. Case Study
Objective: Isolate high-affinity single-domain antibodies targeting B-cell Maturation Antigen (BCMA) from the hundred-billion-scale high-pI library.Background: BCMA, a key member of the TNF receptor superfamily (TNFRSF), plays a pivotal role in plasma cell malignancies and autoimmune disorders. BCMA-targeted therapeutics have achieved major clinical successes, spanning CAR-T cell therapies (e.g., Abecma, Carvykti), ADCs (e.g., Blenrep), and bispecific antibodies (e.g., Tecvayli), all operating via specific binding to BCMA to mediate tumor cell clearance.Methodology: Candidates were identified using the AI-STAL 2.0 system through fully automated library screening, evaluated in silico for developability, and subsequently expressed in eukaryotic systems. Binding affinity to huBCMA-HEK293 cells was evaluated by FACS. Blocking activity against the interaction between BAFF-NFC and huBCMA-ECD-His-Biotin protein was assessed via ELISA.Results: As shown in the figures, FACS binding assays demonstrated that all candidate antibodies exhibited superior affinity compared to the control antibody. Candidates A001, A002, A003, and A004 effectively blocked the binding of BAFF-NFC to huBCMA-ECD-His-Biotin, displaying blocking potency comparable to or superior to the reference antibody.Conclusion: Through AI-STAL 2.0 screening, evaluation, and validation, high-quality VHH candidate molecules with potent cell-binding activity and robust blocking efficacy were successfully identified.
▲ Fig. 8 Affinity testing of antibodies
▲ Fig. 9 Blocking testing of antibodies
V. Conclusion & Outlook
AI-driven antibody design is rapidly transitioning from theoretical proof-of-concept to industrial implementation. Amid this wave, Sanyou Bio's AI-STAL 2.0 High-pI Single-Domain Antibody Discovery System has built a distinct technological moat by integrating three core pillars: full-process AI intelligence, end-to-end automation, and systematic developability evaluation. AI Empowerment Across the Workflow: From library design to candidate screening, AI models are continually refined on massive datasets to guide precision selection. SAI-DA Automated Execution: Automated platforms conduct multi-round panning and high-throughput screening in record time, rapidly covering vast sequence spaces. AI-Driven Developability Screening: Candidates undergo rigorous multi-dimensional predictions (developability, immunogenicity, physicochemical parameters) prior to wet-lab experiments, ensuring only top-tier molecules enter eukaryotic validation.
By uniting a hundred-billion library diversity, high pI (8.0–9.0), zero PTM liabilities, 98% humanization, and a high yield of ~35 lead candidates per project, AI-STAL 2.0 represents a comprehensive methodological shift—moving from random screening to data-guided discovery, from manual operation to automated execution, and from empirical judgment to AI-driven evaluation.
Looking ahead, Sanyou Bio will continue feeding and optimizing its AI models with massive experimental datasets, expanding its antibody library resources and historical data repositories. By deepening the integration between deep learning algorithms and automated platforms, Sanyou Bio aims to drive the end-to-end advancement of single-domain antibody discovery—from sequence design and automated screening to precise developability prediction. AI-STAL 2.0 serves not only as a vast molecular reservoir, but as a high-coverage, highly intelligent, and fully automated discovery engine for novel biologics—empowering global partners to lower technical hurdles, accelerate clinical timelines, and deliver safer, more effective therapeutics to patients worldwide.
推荐阅读
Cyclic Peptide-Generator 环肽从头生成平台
全球招聘 | 三优生物2027年度招聘,火热开启
战略合作 | 三优生物与新码生物达成战略合作
SAI-DA 1.0 | 三优助力下一代ADC
“全程智能”羊驼免疫抗体发现平台
研发前置数字质检,提升湿实验成功率
三优生物原创新药创新工场Clicklinks全新上线
抗体免疫原性风险评估与智能优化
GEOTA智生:开启AI抗体从头设计新路径
2026抗肥胖新靶点全景图:破解“减重不保肌”难题
Peptide-Generator:开启AI多肽从头生成新路径
从180天到14天: 如何重新定义创新药研发速度
AI驱动蛋白质从头设计,加速创新药物发现
兔单抗分子产生系统解决方案
SAI-DA 1.0 | 智优人源化加速器
SAI-DA 1.0 | GEOTA平台
关于三优生物
三优生物是一家以“让天下没有难做的创新药”为使命,以“成为全球创新药研发的引领者”为愿景,专注于创新药的研发及商业转化的高新科技企业。
三优生物以全球领先的智能超万亿分子库 (AI-STAL) 为创新源头,以干湿结合的智能新药加速器 (SAI-DA) 为核心引擎,构建开放式创新系统 (Clicklinks) ,以新药管线驱动,加速创新转化,造福广大病患。
公司总部位于中国上海,在亚洲、北美洲、欧洲等多地建立了业务中心,形成了全球化的业务网络,现有投产及布局的研发及GMP场地20000多平方米。
公司已与全球2000多家药企、生技公司等建立了良好的合作关系。已完成100多个协同研发合作协议,其中10多个创新药进入临床前和临床研究阶段;完成了上千个新药发现及阶段研发项目、数千个品牌产品交付。
公司已申请180多项发明专利,其中30多项发明专利已获得授权,并获得了国家级高新技术企业、上海市专精特新、ISO9001、ISO27001等10余项资质及体系认证。