作者: Kovács, Gábor ; Czakó, Bence ; Sanz, Gema ; Szalai, Bence ; Veres, Daniel ; Gór, Csaba ; Szalay, Kristóf ; Böröcz, Réka ; Mérő, László ; Csendes, Gerold

AbstractPreclinical cancer model systems, such as large-scale drug screens in cancer cell lines, are pivotal for identifying drug-sensitive indications and biomarkers of drug efficacy. However, these systems often lack the tumor microenvironment's complexity, including interactions with immune cells, which play a critical role in modulating drug response. Investigating drug-induced effects on immune cells is experimentally challenging due to the diversity of immune cell types and the complexity of co-culture systems. Computational models of immune cells can help to overcome these limitations and to guide further experiments. Leveraging public post-perturbation single-cell RNA sequencing datasets from various immune cells (including CD8 and CD4 T cells), we trained Turbine’s Simulated Cell engine to predict transcription factor (TF) activity changes in 17 immune cell types post-drug treatment. We compared the predicted, drug induced TF activity changes to immune killing effects from public co-culture experiments. Predicted activity changes of CD8 T-cell activator TFs, such as FOSL and MYC, positively correlated with tumor-killing effects (Pearson r = 0.35 and 0.42, respectively). Conversely, inhibitor TFs such as NR4A1 and IRF4 exhibited inverse correlations (Pearson r = -0.27 and -0.47, respectively).To assess clinical relevance, we analyzed oncology clinical trial data from ACTT. We calculated mean overall response rates (ORRs) for indication - drug pairs and compared it to in vitro indication specific relative IC50 values of drugs. Using a linear model to predict ORRs from in vitro cancer cell line drug response data alone, we predicted ORRs with a Pearson correlation of 0.28. Integrating simulated drug induced TF activities of CD8 T-cells improved this correlation to 0.48, showing that combining tumor-intrinsic mechanisms with immune-mediated cell killing enhances the predictive performance of the model. Our findings suggest that augmenting traditional in vitro drug screening with simulated immune cell activity significantly enhances the clinical translatability of preclinical drug response models, paving the way for improved therapeutic strategies.Citation Format:Gema Sanz, Réka Böröcz, Bence Czakó, Gerold Csendes, Csaba Gór, Gábor Kovács, László Mérő, Kristóf Szalay, Daniel Veres, Bence Szalai. Computational modeling of immune cell phenotypes enhances prediction of clinical anti-cancer drug response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5803.

2025-04-21·Cancer Research

Abstract 2975: Characterization of the RAC1/CDC42 inhibitor MBQ-167 to assess its utility as a tool for interrogating Rho GTPase biology

作者: Veres, Daniel ; Ruiz-Pérez, María Victoria ; Barsi, Szilvia ; Ordasi, Nóra ; Djurec, Magdolna ; Polner-Szundi, Csilla ; Elek, Boldizsár ; Gáspár, Imre ; Fekete, Iván ; Taisz, István ; Varga, Árpád ; Basran, Amrik ; Szántai-Kis, Csaba

The Simulated Cell platform highlighted RAC1 as a target with great antitumor potential. RAC1 is a small Rho GTPase with roles in cell polarity and migration, and contributes to cancer metastasis. To identify good chemical tools for RAC1, we evaluated multiple RAC1 inhibitors, and selected MBQ-167 as a potent RAC/CDC42 inhibitor with publicly available efficacy data.We analyzed the in silico effects of individual and dual RAC1/CDC42 inhibitors using the Simulated Cell. In vitro we studied MBQ-167 using viability, apoptosis and scratch assays. Viability of RAC1 KO cells treated with MBQ-167 was also analyzed. Post-treatment RNA-Seq and bioinformatics analysis of pathway activity signatures revealed MBQ-167 downstream effects. The SignalChem kinase assay was used to validate off-target identification.Our simulations revealed that, beyond their role in cytoskeleton, RAC1 and CDC42 exert effects through apoptosis signaling, with a stronger contribution of RAC1. In vitro, we confirmed the activity of MBQ-167 against RAC1 and CDC42. We observed a negative impact of MBQ-167 on cell migration and an induction of apoptosis. However, RAC1 KO did not affect sensitivity to MBQ-167, suggesting that effects of MBQ-167 on cell viability are not dependent on RAC1.Post-treatment RNA-Seq analysis of MBQ-167 treated cells and calculation of the similarity scores with LINCS consensus signatures, revealed that the MBQ-167 signature is highly similar to those of various kinase KOs and inhibitors targeting PLKs, AURKA, MET, CDK1, MAPK14, FLTs, RET and PDGFRs. RAC KOs were not involved. A kinase off-target panel validated that MBQ-167 inhibits the aforementioned kinases. It inhibited PLK2 and AURKA with an IC50 of around 200 nM, similar to its primary described targets. Other kinases were also inhibited in the nanomolar range, such as CDK1, MET and PLK1.Targeting RAC proteins and CDC42 holds great anti-tumoral potential. We investigated MBQ-167 in vitro and showed that it does not exert its effects on cancer cell viability through RAC1. Besides RAC and CDC42, this compound can inhibit cancer-relevant kinases such as AURKA and PLK2. Simulations helped understanding that MBQ-167 is not a specific inhibitor and therefore not adequate as a tool for interrogating the biology of interest. Further studies are needed to find additional inhibitors against RAC/CDC42 and to understand the effects of MBQ-167 on cancer cells and as a clinical asset tested in Phase I clinical trials as a single agent against advanced breast cancer.Citation Format:Szilvia Barsi, Árpád Varga, Csaba Szántai-Kis, María Victoria Ruiz-Pérez, Iván Fekete, Imre Gáspár, Csilla Polner-Szundi, Boldizsár Elek, Magdolna Djurec, Nóra Ordasi, István Taisz, Amrik Basran, Daniel Veres. Characterization of the RAC1/CDC42 inhibitor MBQ-167 to assess its utility as a tool for interrogating Rho GTPase biology [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2975.

2025-04-21·Cancer Research

Abstract 6502: Inhibition of VCP in an in vivo model of metastatic melanoma selected based on a predictive biomarker demonstrates tumor reduction

作者: Fekete, Ivan ; Szikora, Péter ; Kállai, Dóra ; Elek, Boldizsár ; Connolly, Stephen ; Campbell, Aofie ; Basran, Amrik ; Veres, Daniel ; Djurec, Magdolna ; Hegedűs, Csilla ; Szántai-Kis, Csaba

AbstractValosin containing protein (VCP) is a member of the AAA-ATPase family. It is involved in multiple cellular processes, such as protein degradation, DNA damage response, protein quality control and clearance in the endoplasmic reticulum, all of which are important functions often dysregulated in cancer. VCP is a common essential gene and there have been reports of toxicity associated with its inhibition, therefore a predictive biomarker is needed to stratify patients who are selectively vulnerable. Turbine’s Simulated Cell model highlighted VCP from an in silico target identification screen where a subgroup of melanoma was found to be most sensitive to VCP inhibition (VCPi). These findings were confirmed in a cell line panel covering 12 indications, consisting of 64 cell lines. Next, a simulated drug modifier screen was run using patient relevant avatars with prevalent alterations, which identified MITF as a predictive biomarker of VCPi sensitivity. An in vitro melanoma model of MITF dependency, SKMEL5, was used to test our hypothesis. Hypotheses derived from in silico predictions were validated using various forms of VCP loss including pharmacological inhibition using a phase 1 VCPi, CB5083, as a reference compound, siRNA and by conditional knockout. The sensitivity of VCPi in melanoma was confirmed in vitro and a correlation (R= -0.86) was found between the protein levels of MITF and IC50 values of VCPi. In the SKMEL5 cells, we observed a dose dependent reduction in levels of MITF following VCPi. Next, we sought to test the efficacy of VCPi in the SKMEL5 in vivo CDX model. Four weeks of CB5083 dosing using 50 mg/kg 4on/3off schedule was tolerated, resulted in 46% tumor growth inhibition, and this was accompanied by a downregulation of MITF and one of MITFs targets, DCT, in tumor samples post treatment. These studies support the therapeutic potential of VCP inhibitors in the MITF High/MITF Amplification subgroup of patients with metastatic melanoma with a poor prognosis and a decreased 5-year survival. Targeting a common essential gene such as VCP is associated with a narrow therapeutic index, however, we have demonstrated efficacy using a relatively low dose of 50 mg/kg. Taken together, we are proposing a patient stratification strategy based on the identification of a predictive biomarker from an in silico screen.Citation Format:Aofie Campbell, Dóra Kállai, Boldizsár Elek, Csilla Hegedűs, Magdolna Djurec, Ivan Fekete, Péter Szikora, Csaba Szántai-Kis, Amrik Basran, Stephen Connolly, Daniel Veres. Inhibition of VCP in an in vivo model of metastatic melanoma selected based on a predictive biomarker demonstrates tumor reduction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6502.

项与 Turbine Simulated Cell Technologies Ltd. 相关的新闻（医药）

2025-04-24

·Pharmaceutical Technology

AstraZeneca enters $200m AI cancer pact with Tempus and Pathos

The companies will use the AI-based platform to accelerate cancer drug discovery. Credit: sanjeri via Getty Images. AstraZeneca, Tempus and Pathos AI have signed a multi-year agreement to develop a large-scale multimodal deep learning model designed to accelerate cancer drug discovery. The companies say the foundation model will be used to extract biological and clinical insights, identify novel drug targets, and support the development of new cancer therapies. Under the terms of the deal, Tempus will receive $200m in data licencing and model development fees. The company will contribute its large library of de-identified oncology data to help build the model, which will be shared by all three parties once completed. Tempus’ shares, listed on the Nasdaq exchange, rose 14.6% on 23 April following the announcement, closing at $49.55, up from $43.23 the previous day. “Generative artificial intelligence (AI) and the emergence of large multimodal models is the final catalyst needed to usher in precision medicine in oncology at scale,” said Tempus’ CEO Eric Lefkofsky. “Tempus has spent the last decade investing billions of dollars into collecting the necessary data needed for a foundation model of this kind to take shape.” The partnership builds on an existing relationship between Tempus and AstraZeneca. The two companies first teamed up in 2021, announcing an AI-driven oncology R&D alliance. In 2023, they expanded the partnership with the deployment of Tempus’ Next platform to support clinical decision-making for non-small cell lung cancer (NSCLC) patients, helping physicians determine whether patients should undergo guideline-directed molecular testing. GlobalData Strategic Intelligence US Tariffs are shifting - will you react or anticipate? Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis. By GlobalData Learn more about Strategic Intelligence AstraZeneca is also advancing its use of AI across multiple areas of drug development. “Cancer drug discovery and clinical development are being transformed by the ability to analyse vast amounts of rich data using artificial intelligence,” said Jorge Reis-Filho, chief AI and data scientist of oncology R&D at AstraZeneca. The company has pursued other AI collaborations as well, including a deal with biological simulation company Turbine. Announced in January 2024, the partnership uses AI-driven simulations to explore drug resistance mechanisms in blood cancers, aiming to predict drug interactions and identify potential combination therapies. Unlike traditional lab-based methods, these models can generate millions of simulations to analyse complex biological pathways at a fraction of the time. AstraZeneca’s senior director in oncology data science Krishna Bulusu recently shared that AI is transforming drug discovery, but its implementation must be thoughtful and strategic. He said at the ELRIG Drug Discovery meeting in London on 11 March 2025 that AI must be a “thought partner” in drug discovery. The use of AI in drug development continues to gain momentum across the pharma industry. According to a survey by GlobalData, the parent company of Pharmaceutical Technology , AI is considered the most disruptive technology among businesses, including in the healthcare industry. Earlier in April 2025, Eli Lilly entered a research partnership with BigHat Biosciences to co-develop antibody therapeutics using BigHat’s AI-based Milliner platform. BigHat has also signed agreements with Johnson & Johnson (J&J), AbbVie, MSD, and Amgen.

2025-03-14

·Pharmaceutical Technology

AstraZeneca director says AI must be a “thought partner” in drug discovery

AstraZeneca’s Krishna Bulusu presented at the ELRIG Drug Discovery conference in London on 11 March. Credit: SOPA Images via Getty Images. Artificial intelligence (AI) is transforming drug discovery, but its implementation must be thoughtful and strategic, according to Krishna Bulusu, senior director in oncology data science at AstraZeneca. Speaking at the ELRIG Drug Discovery meeting in London on 11 March 2025, Bulusu outlined how AI can improve efficiency, reduce costs, and support personalise medicine, but cautioned that its success depends on data integration, model explainability, and context-specific predictions. “Accelerating drug discovery doesn’t just mean doing the same thing very, very fast. It means that we’re also going to do different things, and we’re going to do things more efficiently,” Bulusu said. He stressed that AI should not be applied indiscriminately but should be used to answer well-defined scientific questions. One key example of how AstraZeneca is using AI role in drug discovery is the company’s collaboration with biological simulation company Turbine. The partnership, announced in January 2024, uses Turbine’s simulated cancer cell technology to model drug resistance mechanisms in haematological cancers. By integrating public and proprietary data, the AI-driven model generates millions of simulations, predicting drug interactions and identifying potential combination therapies. “Now that is powerful, right?” Bulusu asked. “Because the scalability aspect – if I have to go to the lab and do this, it’ll cost me a lot of money.” Unlike traditional lab-based research, which can be time-consuming and expensive, AI simulations allow researchers to explore complex biological pathways at a fraction of the time. The model provides quantitative insights into how different inhibitors impact cellular pathways, offering a new way to generate hypotheses and refine drug development strategies. “From a big pharma perspective, this is great because I have a portfolio of drugs, and you’re telling me where to position my portfolio drugs, either as normal therapy or in combinations. That’s why we work very closely with Turbine on this,” Bulusu added. The idea of AI-driven biological simulation is gaining traction across the industry. Demis Hassabis, CEO of DeepMind and co-creator of AlphaFold, has previously shared his vision for AI-powered virtual cells: “My dream is to eventually have virtual cells, like a simulation of a virtual cell. We’re maybe ten years away from that,” Hassabis said at the Financial Times Pharma and Biotech Summit in November 2024. However, Bulusu underscored that AI’s success in drug discovery is not just about technological advancement but also about ensuring that models are interpretable, and predictions are relevant. “The value for AI, at least with the larger organisations in the past, is when a non-data scientist understands and appreciates them. And this doesn’t happen unless you’re working together.” Investigating the scope of AI in drug discovery Despite AI’s potential, significant challenges remain. Bulusu pointed to gaps in longitudinal patient data, biases in AI models, and the need for better early disease detection. “AI needs to become a thought partner. And for that to happen, the confidence and trust in what we’re doing as data scientists needs to come through,” he said. To address these challenges, Bulusu emphasised the need for improved data collection and integration, particularly capturing the full patient journey from preclinical to clinical stages. “We are very bad as a community at generating longitudinal data,” he said. Longitudinal data needs to be colletcted before preclinical and clinical research, but this is currently happening in reverse. This is important, because Bulusu said while “a patient’s journey is very important to capture, we’re just taking snapshots of it.” Looking ahead, Bulusu emphasised the importance of starting with the right scientific question before selecting an AI model. “From an AI perspective, value and impact is and will always be driven by starting with the right question and then asking, what’s the right model to answer that question. It’s never the other way around,” Bulusu concluded.

2025-01-24

·智药邦

Nat Cancer｜人工智能对于癌症药物研发的作用

2024年12月17日，Nature Cancer上发表文章Beyond the hype of AI in cancer R&D。以下是全文。人工智能有望更快、更廉价、更精准地助力药物制造。尽管由人工智能驱动的生物技术公司吸引了大笔资金，还登上了耀眼的头条新闻，但它们尚未将产品成功推向市场终点。到目前为止，人工智能的最大影响体现在何处呢？由人工智能驱动的生物技术公司宣称要“重新思考药物发现”以及“实现研发产业化”，这成功吸引了数十亿美元的资金投入。2024年4月，ARCH孵化的Xaira公司筹集了破纪录的10亿美元，用于部署其人工智能辅助药物设计工具套件。总体而言，以人工智能为核心的生物技术公司在2024年所筹集的风险资金，有望超过2021年以来的任何一年。随着所有运用复杂计算的公司都在寻求搭乘人工智能的发展顺风车，各公司的宣传内容也在不断更新。这种兴奋是有充分理由的。从靶点发现、假设生成和药物设计，到生物标记物识别、试验设计以及患者招募，人工智能有潜力加速并改善生物医学的几乎每一个环节。鉴于有大量“组学”技术、成像及其他患者数据可用于机器训练，癌症研发领域自然非常适合大规模应用人工智能。然而，人工智能并非仅仅是几十年来以各种形式应用于生物技术领域的高能数据处理。如今的算法能够大规模地分析和处理数据，挖掘出以前未知的联系，产生并测试新想法，还能助力确定生物途径和机制。拥有足够的训练数据后，最强大的人工智能引擎能够生成新的分子结构，这些结构可单独或组合使用，以有效治疗疾病。2024年5月发布的谷歌DeepMind蛋白质结构数据库AlphaFold3，涵盖了预测蛋白质如何与核酸等其他分子结合的内容。图2 AF3能够准确预测生物分子复合物的结构细胞生物学初创公司Cellanome的联合创始人、液体活检平台GRAIL（今年被Illumina剥离）的前任联合创始人Mostafa Ronaghi总结：“随着定量生物学的兴起，人工智能是这场游戏中的关键调味品。” 然而，尽管有几种人工智能候选药物正在进行临床试验，但还没有一种人工智能候选药物通过监管机构的审批。第一代人工智能驱动的药物发现公司，如BenevolentAI和Exscientia（2024年8月被竞争对手Recursion兼并），均曾遭遇挫折。 Lantern Pharma公司CEO Panna Sharma表示：“人工智能公司有很多‘学习很酷东西’的活动，但很少能看到公司充分利用人工智能并将药物推进到有意义的临床试验阶段。” Insilico Medicine和Recursion的非肿瘤生成式人工智能项目，最近公布的II期数据很有希望，但因缺乏足够的统计细节，所以难以表明人工智能明显领先于传统方法。REC-994是一种治疗症状性脑海绵状畸形的药物，其II期试验数据显示该药物安全且可耐受，但12个月后的疗效有限，Recursion的股价也随之下跌。即便如此，Recursion首席研发官兼首席商务官、强生前任首席数据科学官Najat Khan表示：“我们并不期望任何技术的首次迭代就是完美的。我们正在寻找具有潜力的早期迹象。” 迄今为止，人工智能对医疗保健最显著的影响体现在诊断成像方面。机器的模式识别能力有助于提高乳腺癌的早期发现率，能够突出显示异常情况，或标记检查结果以便后续跟进。FDA批准的大多数人工智能工具都与成像相关，尽管在广泛应用和临床转化方面存在挑战，但人工智能驱动的诊断工作仍在持续拓展。例如，数字病理学公司Owkin最近与默克（Merck MSD）合作开展了一项试点计划，试图借助分析肿瘤组织的机器学习（ML）算法，提高结直肠癌诊断的准确性。聚焦于药物发现阶段在幕后，人工智能工具正在变革研发工作。但是，自第一家人工智能公司诞生以来的十多年间，大部分工作集中在早期发现阶段，而这一阶段较难吸引大众关注、获取头条新闻。 Sharma说到：“这项技术能够缩短从早期发现到确定明确候选药物的时间。与传统方法相比，Lantern的癌症项目从最初的人工智能洞察到首次人体临床试验的时间缩短了约一半，成本降低了80%。成本约为100万到250万美元，耗时仅两到三年，而不是通常的四到七年。”Recursion和Insilico Medicine的情况也与之类似。 Insilico Medicine创始人兼首席执行官Alex Zhavoronkov表示：“对于一个中等新颖的靶点，从构思到临床前候选研究平均需要13个月，而开展人体临床试验则需要20-24个月。” 这种速度得益于人工智能能够生成数十种可行的候选药物，并大规模地进行多次实验和模拟。以癌症为重点的Turbine首席执行官Szabolcs Nagy说：“你可以测试更多的想法。先导优化--使化合物更接近药物的状态，是研发中的一大难题，而人工智能可以帮助解决这一问题。”Turbine的细胞模拟平台利用机器学习建立人体细胞组织的计算机模型，使开发人员能够在进行湿实验室实验之前，观察特定治疗方法的效果。机器虽然还无法取代实验室测试，但它们能让研究人员更快地开展正确的实验，从理论上讲，这有助于提高日后的成功率。再利用和组合人工智能对癌症治疗的一些早期影响，可能体现在对失败或废弃药物的重新利用上。Lantern有三个临床项目，分别涉及非小细胞肺癌（NSCLC）、晚期实体瘤和非霍奇金淋巴瘤。先导药物是一种II期二硫化物，可在某些从不吸烟的肺癌患者中增强酪氨酸激酶抑制剂的作用，它是一种重新利用的药物，是通过Lantern的人工智能平台确定的。该平台整合了来自研究、试验和数据库的数百亿肿瘤数据点，并运用机器学习策略预测患者对候选药物的反应。 Sharma说：“人工智能可用于快速发现被忽视的新适应症，或识别尚未完全明确的新癌症亚型（含生物标志物）。” 同样，Recursion的精准肿瘤学管线，由武田一项已停产的NSCLC资产打头阵：一种MEK1和MEK2的异位小分子抑制剂，目前正快速应用于晚期AXIN1或APC突变癌症。为了更好地了解该分子的作用机制，Recursion利用机器学习算法将其与各种疾病和细胞模型进行交叉对比，从而确定了该药物的最佳作用位点。针对AXIN1或APC基因突变的晚期或转移性癌症患者的II期试验正在进行中，预计将于2025年获得初步数据。 Recursion产品管线确定抗癌药物的有效组合，是另一个容易取得成果的领域。Sharma说：“人工智能使我们能够“浏览所有试验数据，找出应该联合使用的药物”。当前的联合用药试验既艰巨又耗时，有时sponsor之间的利益分歧也会阻碍试验的进行。人工智能能够更快地提供更可靠的证据基础。Turbine利用其细胞模拟平台，为Debiopharm的I期WEE1抑制剂Debio 0123寻找新的组合疗法，帮助确定并验证了治疗搭档cabozantinib（Exelixis的酪氨酸激酶抑制剂）。” 快速开发新化合物人工智能能够为特定治疗效果量身定制新结构，这一点令一些支持者最为兴奋。在Zhavoronkov看来，这种“生成”能力将“真正的”人工智能与机器学习区分开来，机器学习中计算机只是学会遵循特定指令，无法超越指令的范畴。“真正的人工智能是深度学习--这意味着能够进行推理和处理多模态数据，创造出原本不存在的东西，而且比人类做得更好。在生物科学领域之外，一个备受瞩目的例子是谷歌的AlphaGo，它在没有经过专门针对围棋的明确训练的情况下，就学会了下围棋，并且足以打败人类专家。” 如今，人工智能生成的生物分子与现有的生物分子较为相似，都是经过调整，以提高选择性或降低脱靶毒性。考虑到算法训练数据的特性，这种情况并不意外。Insilico Medicine用于治疗BRCA突变型癌症的USP1抑制剂，源自该公司的小分子生成人工智能平台Chemistry42。USP1是一种调节DNA损伤反应途径的去泛素化酶，并非新靶点。但这种抑制剂具有新颖的结构，而且其临床前研究内容，包括抗肿瘤活性、耐受性和药代动力学，足以吸引许可合作伙伴Exelixis在2023年预付8000万美元，目前该药物正在进行I期测试。Tango Therapeutics的一个类似候选药物，因肝毒性于2024年5月被放弃。罗氏也在寻找一种有效的USP1抑制剂，以补充另一种合成致死靶点PARP的抑制剂。 Insilico Medicine三项处于临床阶段的癌症资产中的另一项，是利用该生物技术公司的靶点识别平台PandaOmics设计的QPCTL抑制剂。QPCTL与CD47-SIRPα轴相关，有助于癌症躲避免疫监视。该化合物与上海复星医药合作，于2024年5月在中国进入1期剂量递增试验，用于治疗CD47-SIRPα高度参与的淋巴瘤和晚期实体瘤。 Recursion唯一处于临床阶段的新癌症候选药物，是Exscientia精确设计的CDK7抑制剂，目前正处于1/2期试验阶段，用于治疗对标准疗法无反应的实体瘤患者。CDK7是调控癌基因转录和细胞周期进展的关键因素，是另一个研究较为深入的靶点。目前正在进行的1b/2期试验，包括Carrick Therapeutics的samuraciclib，这是一种CD47抑制剂，目前正与一种口服PROTAC雌激素受体降解剂联合治疗ER阳性、HER2阴性转移性乳腺癌患者，这些患者之前曾接受过CDK4/6抑制剂治疗。其中一些竞争药物在争取足够广泛的治疗窗口期方面一直面临困难。Recursion/Exscientia设计的小分子药物具有严格控制的结合范围和持续时间，这有望使其在同类产品中脱颖而出。对于一流的癌症资产，Khan指出Recursion的RBM39强效选择性降解剂，RBM39是肿瘤相关基因表达的调控因子，在许多癌症中均有富集。该公司计划在2024年申请研究性新药批准，并于次年开始在生物标志物筛选出的实体瘤或淋巴瘤患者中进行试验。试验设计和执行先导候选药物必须在人体中证明其有效性。临床试验耗费了大部分研发成本和时间，因此即使是微小的疗效提升，也会产生巨大的影响。“临床开发中最大的挑战之一，就是我们没有针对合适的患者。”Khan说。利用来自Tempus等专业数据采集器的临床和多组学数据，Recursion部署了因果人工智能模型，以识别最佳受试者。改进患者选择意味着可以缩小试验规模，理论上也能提高成功率。人工智能--这里指的是标准的数据运算迭代，而非更具创新性的生成版本，还可用于定位符合条件的患者，并确定最佳试验地点，以最大程度加快招募进程。Khan表示，大型制药公司多年来一直在试验运营中部署人工智能（诺华于2018年推出了试验监测中心），这提高了招募率和多样性。进化，而非革命人工智能的支持者承认，人工智能的强大威力尚未得到充分展现。它加快了研发的某些环节，但还未能比药物化学家和生物学家更快地提供更优质的药物。迄今为止，人工智能对药物研发的影响是渐进式的，而非革命性的。这在一定程度上是因为人工智能的价值，来源于与其他技术的融合，以及人类的专业知识和精心策划。将人工智能工具融入传统研发流程并非易事。大型制药公司面临着内部阻力，员工担心会因此失去工作。规模较小的人工智能公司则缺乏资金来升级和重建整个流程，它们必须努力证明自己的解决方案切实可行。随着人工智能的广泛应用，不同的商业模式也在不断涌现（见box 1）。人工智能揭示疾病复杂性的能力，仍然受到输入数据质量的限制，这些数据比人工智能所带来的兴奋感所暗示的更杂乱、更不完整、更不全面。Cumulus Oncology的首席数据科学家Phil Chapman说：“大量的健康数据确实存在，但其中大部分数据仅来自少数人，而且往往复杂、混乱，也不是纵向的。”监管机构也必须跟上步伐：FDA最近成立了一个人工智能监督委员会，以指导如何在监管申报中考虑人工智能因素，但详细的政策似乎仍需时日才能出台。尽管面临诸多挑战，但Zhavoronkov相信，在本十年内，人工智能对药物研发的影响将出现令人信服的证据。他表示，这将涉及展示“从靶点发现到药物设计的一条清晰、有据可查的途径”，以及提供能够造福患者的重磅疗法。“如果没有战争或经济崩溃等重大变故，我预计这将在2026年至2029年之间实现。” 同时，随着人工智能持续渗透到各个领域和活动中，“当然会存在炒作现象。”他补充道。 Box 1 端到端人工智能与聚焦人工智能随着人工智能在研发领域的广泛应用，各种商业模式不断涌现。Xaira、Insilico Medicine和Recursion等公司，试图将从药物研发构思到获批的整个流程进行高端化改造，研究多种疾病和治疗假设，并建立自己的研发管线来验证自身方法。Insilico Medicine还通过向制药公司授权其人工智能平台来获取收入。其他一些公司则专注于研发的特定环节，如小分子发现（Atomwise或Iktos）、蛋白质设计（Cradle）或试验招募。法国初创公司Cure51正在建立一个癌症幸存者数据库，并利用人工智能寻找这些人的共性，从而揭示生存机制，为治疗提供灵感。人工智能还被用于创建细胞、蛋白质组和分子相互作用的基础图谱和模型，以揭示生物的复杂性，并提供加速研发的工具。Turbine的团队利用机器学习网络来反映、理解和预测蛋白质的相互作用，从而建立了一个适用于任何细胞的人类蛋白质逻辑模型。相关人员表示，这一模型可应用于药物发现以外的领域，一直到临床概念验证阶段，在任何需要决定“选择哪种分子、哪种剂量或哪种适应症”的情况下都能发挥作用。Cellanome的目标是建立细胞和细胞相互作用的基础模型，将基因组学、蛋白质组学、代谢组学等多模态测量数据纳入其中，并进行长期采集。Owkin正在收集病人的多模态数据，以生成分子空间环境图集，绘制肿瘤微环境图以及肿瘤与免疫细胞相互作用的图谱。参考资料： Senior, M. Beyond the hype of AI in cancer R&D. Nat Cancer 5, 1759–1761 (2024). https://doi.org/10.1038/s43018-024-00857-3 --------- End --------- 感兴趣的读者，可以添加小邦微信加入读者实名讨论微信群。添加时请主动注明姓名-企业-职位/岗位或姓名-学校-职务/研究方向。

100 项与 Turbine Simulated Cell Technologies Ltd. 相关的药物交易

登录后查看更多信息

100 项与 Turbine Simulated Cell Technologies Ltd. 相关的转化医学

登录后查看更多信息