PO.CH01.02 · 化学
一个集成AI的转化筛选平台，用于临床预测性地发现安全有效的巨胞饮依赖性癌症疗法
A translational screening platform with AI integration for clinically predictive discovery of safe and effective macropinocytosis-dependent cancer therapeutics
该海报暂无可下载的资料
AACR 官方页面
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：巨胞饮依赖性癌症依赖营养清除通路以在应激条件下维持肿瘤生长。目前的疗法很少能提供持久的控制，这凸显了将癌症进展稳定为慢性疾病的策略的必要性。尽管已批准药物存在大量临床数据——将作用机制（MoA）、安全性和疗效联系起来——但临床可转化的、与巨胞饮相关的基于细胞的筛选仍然有限。我们开发了一个AI驱动的转化发现平台，将表型筛选与临床关联相结合，以识别安全、机制上有意义且具有临床预测性的候选疗法。
方法：优化了一种自动化的基于图像的细胞计量分析方法，以最小的人为偏倚定量癌细胞中的巨胞饮。使用这种中通量方法筛选了FDA批准的药物和专有化合物。AI算法将体外表型与回顾性临床数据相关联，整合了MoA、毒性和疗效指标。机制聚类识别了临床有效药物之间共有的治疗通路。AI建模还预测了临床表现，并生成了强调安全性和转化潜力的合理联合用药假设。
结果：AI验证证实了表型结果与来自892种FDA批准的癌症及非癌症药物的临床安全性/疗效特征之间的高度一致性，展示了高转化保真度。机制聚类揭示了一条与持久获益和低毒性相关的汇聚信号轴。若干新型自主研发化合物表现出可比的抑制特征和AI预测的临床潜力。AI建模提出了已批准药物与具有未充分探索机制的专有药物之间的协同组合，优先考虑安全、持久的治疗反应。
结论：该平台将无偏倚的表型分析与AI引导的临床验证相结合，能够早期识别具有治疗意义且可转化的候选药物。发现共有的机制靶标和AI预测的新型化合物，展示了一个以临床为基础、数据驱动的模型，用于开发安全、有效、面向慢性疾病的癌症疗法。
关键词：表型筛选；巨胞饮；AI药物发现；转化肿瘤学；临床验证；机制聚类；联合治疗；药物再利用；安全性分析
查看英文原文 English abstract
Background: Macropinocytosis-dependent cancers rely on nutrient-scavenging pathways to sustain tumor growth under stress. Current therapies rarely provide durable control, underscoring the need for strategies that stabilize cancer progression as a chronic condition. Although extensive clinical data exist for approved drugs-linking mechanisms of action (MoAs), safety, and efficacy-clinically translatable, macropinocytosis-relevant cell-based screens remain limited. We developed an AI-driven translational discovery platform integrating phenotypic screening with clinical correlation to identify safe, mechanistically meaningful, and clinically predictive therapeutic candidates.
Methods: An automated image-based cytometry assay was optimized to quantify macropinocytosis in cancer cells with minimal human bias. FDA-approved drugs and proprietary compounds were screened with this medium throughput method. AI algorithms correlated in vitro phenotypes with retrospective clinical data, integrating MoA, toxicity, and efficacy metrics. Mechanistic clustering identified shared therapeutic pathways among clinically effective agents. AI modeling also predicted clinical performance and generated rational combination hypotheses emphasizing safety and translational potential.
Results: AI validation confirmed strong concordance between phenotypic results and clinical safety/efficacy profiles from 892 FDA-approved cancer and non-cancer drugs, demonstrating high translational fidelity. Mechanistic clustering revealed a convergent signaling axis associated with durable benefit and low toxicity. Several novel in-house compounds exhibited comparable inhibition profiles and AI-predicted clinical potential. AI modeling proposed synergistic combinations between approved and proprietary agents with underexplored mechanisms, prioritizing safe, durable therapeutic responses.
Conclusions: This platform unites bias-free phenotypic assays with AI-guided clinical validation, enabling early identification of therapeutically meaningful and translatable drug candidates. Discovery of a shared mechanistic target and AI-predicted novel compounds demonstrates a clinically grounded, data-driven model for developing safe, effective, and chronic disease-oriented cancer therapies.
Keywords: Phenotypic screening; macropinocytosis; AI drug discovery; translational oncology; clinical validation; mechanistic clustering; combination therapy; drug repurposing; safety profiling
利益披露 Disclosure
Q. Chu, None..
A. Bonanno, None..
Y. Kong, None.