Research and Resources

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.

This study demonstrates Zephyr AI’s multi-modal machine learning method for identifying late-stage ovarian cancer patients likely to respond to olaparib, independent of HRD status. Our model significantly outperformed conventional biomarker approaches, improving real-world survival outcomes and offering biological insights through our Vulnerability Networks™

Zephyr AI’s generative Bayesian network approach synthesizes comprehensive molecular profiles from limited NGS panel data in lung and breast cancer. This method enhances sparse tumor data to enable downstream analysis, biomarker discovery, and advanced ML model development without additional testing, bridging a major gap in real-world precision oncology.

Zephyr AI’s Mut2Ex model reconstructs tumor gene expression profiles using only genetic data from commercial NGS panels combined with minimal clinical information. Trained on ~1,200 DepMap cell lines, Mut2Ex generated expression profiles for ~10,000 TCGA and ~180,000 GENIE tumors, substantially enhancing the clinical utility of real-world clinicogenomics data for precision medicine applications.

Zephyr AI developed a machine learning model that reconstructs tumor gene expression from NGS panel data to predict tumor microenvironment (TME) features critical for immune checkpoint inhibitor (ICI) response. Trained on over 8,000 tumors across 32 cancer types, the model delivers high reconstruction accuracy and supports personalized immunotherapy strategies using real-world clinical data.

This research introduces ImmunoBERT, Zephyr AI’s machine learning model that predicts survival in cancer patients receiving immune checkpoint inhibitors. Integrating drug structure, target binding profiles, and NGS panel data from 1,700 patients, ImmunoBERT outperformed top DREAM challenge submissions and reconstructs key tumor microenvironment features, enabling personalized immunotherapy selection without specialized testing.

Zephyr AI developed a systems biology-informed machine learning model to predict sensitivity to Tagrisso (a third-generation TKI) in NSCLC using clinically available NGS data. Validated across NSCLC cell lines, PDX models, and a real-world cohort (n=334), the model identifies pharmacological vulnerabilities and reveals a potential opportunity for label expansion beyond current FDA indications.

Zephyr AI, in collaboration with MedStar Health, developed interpretable AI/ML models that predict 0–5 year risk of adverse outcomes in type 2 diabetes using real-world EHR, claims, and social determinants data from over one million patients. The models deliver strong performance (AUC ~0.85) and provide actionable, transparent predictions to support earlier clinical interventions.