ADC Target Profiling in NSCLC: Generalizable AI Separates TROP-2 and cMET Phenotypes.

Area of Science:

• Computational pathology
• Biomarker discovery
• Artificial intelligence in oncology

Background:

• Antibody drug conjugates (ADCs) targeting TROP-2 and cMET are emerging therapies for non-small cell lung cancer (NSCLC).
• Accurate biomarker assessment is crucial for ADC efficacy but is currently limited by subjective visual scoring and poor reproducibility.

Purpose of the Study:

• To develop and validate a modular AI pipeline for objective and reproducible quantification of cancer biomarkers in NSCLC.
• To assess the generalizability of the AI scorer across different biomarkers, including TROP-2, cMET, HER2, and PD-L1.
• To integrate AI-driven expression maps with clinicopathologic, molecular, and tumor microenvironment (TME) features to identify novel therapeutic strategies.

Main Methods:

• A modular AI pipeline was developed to detect cells, classify carcinoma cells, and quantify membranous and cytoplasmic protein expression.
• A scorer trained on TROP-2 was applied zero-shot to cMET, HER2, and PD-L1.
• The pipeline analyzed 1,142 resected NSCLC samples, integrating expression data with comprehensive clinical, molecular, and TME features.

Main Results:

• The AI scorer demonstrated high concordance with pathologist annotations for TROP-2 (r=0.98-0.99) and generalized effectively to cMET, HER2, and PD-L1.
• AI performance closely matched interobserver variability among six pathologists.
• Expression mapping revealed distinct TROP-2 (high in LUSC, associated with immune-deserted TME) and cMET (high in LUAD, associated with immune-active TME and KRAS mutations) phenotypes.

Conclusions:

• Foundation model-based AI scoring provides expert-level, scalable biomarker quantification for NSCLC.
• Distinct TME phenotypes associated with TROP-2 and cMET expression suggest potential therapeutic strategies, such as combining ADCs with immunotherapies or kinase inhibitors.