At this year’s ESMO Congress in Berlin, one message was clear: AI-based biomarkers are exiting the lab and entering real clinical decisions. From digital pathology to CT radiomics and multi-omics, teams showed practical ways to read response earlier and adapt treatment more precisely.

What stood out

• Digital pathology in pMMR mCRC to enrich for benefit from adding immunotherapy.

• Mesothelioma workflows pairing objective CT volumetry with genomic heterogeneity signals.

• Resectable NSCLC using radiomic deltas, optionally combined with early on-treatment ctDNA, for pCR prediction and prognosis.-

• ALK+ NSCLC applying AI to early lesion-level responses (including brain metastases) to anticipate PFS better than first follow-up RECIST.

Why this matters

Across indications, the pattern is consistent: AI makes “response” measurable sooner, from pixels and slides to circulating and genomic signals, and ties that measurement to outcomes. That’s what a clinical biomarker should do: operate on routine data, predict something that changes care, and help patients avoid ineffective therapy and unnecessary toxicity.

Our take

To make this everyday medicine, the field needs:

• Prospective, pre-specified validation with locked models and clinical-utility endpoints.

• Site-agnostic robustness across scanners, stains, protocols, and populations, with transparent calibration and drift monitoring.

• Human-in-the-loop, auditable workflows that surface case-level explanations and uncertainty, no black boxes at tumor board.

• Regulatory-grade operations treating AI biomarkers like assays: versioned releases, change control, and post-market surveillance.

What we’re building

At Nexomic, we’re pioneering multimodal AI biomarkers by integrating multi-omics into a single, calibrated patient-state signature. The result: precise treatment-benefit prediction and earlier molecular response calls.

Bring Nexomic into your next trial, multi-omic AI that moves the needle.