机器学习加速发现协同性PI3K-mTOR组合，用于强效抑制KRAS G12突变患者来源异种移植瘤的肿瘤生长

Identifying effective drug combinations remains a major challenge in oncology, primarily due to inherent resistance and the limited clinical fidelity of conventional in vitro screening data. To address this translational barrier, we established a robust platform utilizing Certis Oncology's proprietary patient-derived xenograft (PDX) models, which offer high correlation with clinical responses, to predict and validate synergistic therapeutic regimens. Our approach leveraged CertisAI TM , an ensemble of proprietary machine learning (ML) models trained on high-throughput combination experiments. CertisAI TM successfully prioritized combinations targeting the PI3K and mTOR pathways as lead candidates for treating aggressive KRAS G12 mutant tumors, spanning non-small cell lung, gastric, pancreatic, and colorectal cancers. We experimentally evaluated a comprehensive matrix of six PI3K inhibitors and four mTOR inhibitors across distinct KRAS G12 mutant PDX models. These ex-vivo studies rigorously demonstrated that the dual inhibition of the PI3K-mTOR axis resulted in potent, synergistic anti-tumor activity, consistently yielding efficient and sustained tumor cell reduction across all tested models. Notably, a key synergistic combination-involving Everolimus or Tacrolimus (mTORi) paired with Inavolisib (PI3Ki)-translated directly to significant therapeutic benefit in the PDX setting. These findings validate our PDX-informed strategy for rapidly identifying clinically relevant, synergistic combinations. The demonstrated efficacy of the Everolimus/Tacrolimus + Inavolisib combination provides compelling preclinical evidence, establishing this dual-agent approach as a strong therapeutic candidate for clinical evaluation in KRAS G12 mutant solid tumors.