Who Actually Responds? Multi-Omics Patient Stratification on BioMate

Most drugs work for some patients and not others. Finding the "some" before the trial is patient stratification — the single highest-leverage move in modern drug development. Our Patient Stratification survey maps the whole field; here's how BioMate puts it to work.

The one distinction that matters

The throughline is predictive vs. prognostic: a prognostic marker tells you the likely outcome regardless of treatment; a predictive marker tells you who benefits from a specific therapy — and only the latter licenses a treatment decision. Isolating the predictive signal takes multi-omics factor models, causal methods, and enrichment designs — not plain outcome prediction.

What BioMate does

• Cross-modal responder stratification. BioMate runs MOFA2 joint decomposition across RNA-seq + Olink proteomics on the same patients. In a worked CIDP analysis, a complement factor separated responders at AUROC 0.79 and an FcRn factor at 0.71 — and the pipeline reproduced the canonical CLL benchmark (IGHV status, AUROC 0.98) first, so you trust the harness before the result.
• Single-cell responder biology. scRNA-seq clustering → responder differential expression → pathway enrichment in one session (e.g., an anti-TNF non-responder signature with CXCL13 +4.2 log₂FC).
• Molecular subtyping. Transcriptomic subtype classifiers and interferon-signature scoring connect a cohort to the treatment-relevant subgroup — the logic behind consensus molecular subtypes and IFN-high trial enrichment.
• Reproducible scoring. Every stratifier is scored against gold-standard labels with a documented metric harness — the validation the field's own literature says most published models skip.