Multimodal AI predicts immune checkpoint inhibitor response from clinically available inputs and whole-slide images with explainable tumor biology and combination therapy insights

This research presents Zephyr AI’s AIM-io model, a multimodal machine learning approach that predicts response to immune checkpoint inhibitors using routine clinical inputs, across 3 independent datasets: tissue-derived data, liquid biopsy, and whole-slide images. Across multiple real-world cohorts, AIM-io improves prediction of survival outcomes compared to conventional biomarkers and identifies biologically meaningful tumor microenvironment features associated with response. The model reconstructs immune programs and therapeutic vulnerabilities, enabling interpretable insights into mechanisms of sensitivity and resistance. These capabilities support more precise patient stratification and rational design of combination immunotherapy strategies.

Real world prediction and biological characterization of sotorasib sensitivity using multimodal AI and liquid biopsy genomic inputs

This study demonstrates Zephyr AI’s multimodal AIM-Bx platform for predicting response to KRAS inhibitors in non-small cell lung cancer using clinically available data, including ctDNA. Applied to a real-world cohort, the model identifies patients deriving meaningful benefit from sotorasib beyond KRAS mutation status alone. AIM captures underlying tumor states and pathway dependencies associated with response heterogeneity, including MAPK reactivation and alternative signaling programs. These findings highlight the value of functional, biology-informed biomarkers for patient stratification and therapeutic decision-making in KRAS-driven cancers

Identifying Novel Drivers of Drug Sensitivity Using an AI-Enabled Multi-Modal Biomarker – Osimertinib Sensitivity Beyond EGFR

This research highlights Zephyr AI’s AIM-Bx platform—a multi-modal, AI-enabled biomarker that predicts response to osimertinib in NSCLC beyond traditional EGFR mutation status. Trained on routinely available clinical and genomic data from tissue and liquid biopsy, AIM-Bx identified EGFR+ non-responders and EGFR– responders, uncovering transcriptomic programs and tumor dependencies not captured by DNA-level biomarkers. Predictions were validated across diverse real-world datasets (n=893), demonstrating significantly improved survival in predicted responders. These findings support AIM-Bx as a clinically adaptable, biologically grounded tool for broader patient stratification and AI-enabled CDx development.