Revolutionizing drug discovery in lung cancer: An artificial intelligence (AI)-assisted framework for identifying target antigens for antibody-drug conjugates

Abstract

Identifying appropriate target antigens continues to be a hindrance in the development of antibody-drug conjugates (ADCs), particularly for lung adenocarcinoma (LUAD). This paper presents a rule-based, Artificial Intelligence (AI)-assisted system that automates the processes of data harmonization, filtering, and prioritization within extensive transcriptome datasets. The TCGA-LUAD (20,530 genes) and GTEx lung (57,233 genes) datasets were harmonized; protein-coding, surface-localized molecules were evaluated for differential expression (Δlog2), housekeeping or essentiality, projected internalization, solubility, and subcellular accessibility. We ranked the candidate molecules by using a composite score, combining normalized Δlog2 and internalization category. We then curated the ranked molecules for their function, relevance to cancer, tissue specificity, and translational feasibility. These processes results in 647 high confidence surface molecule candidates. Several recognized ADC targets (CEACAM5, MET, LRRC15, MUC16) were included in this list, supporting internal validity. Five antigens (PROM2, DSG2, SEZ6L2, CDH3, and CDCP1) met the quantitative thresholds and translational criteria, with commercial antibodies available for testing. Thus, this reproducible, scalable workflow may reduce subjective bias, clarify decision logic, and offer a general template for antigen discovery in the oncology settings. We believe that this combination of scalability, automation, standardization and validation represents a substantial step compared to conventional expert-curated, manually filtered workflows.