Abstract
<title>Abstract</title> <p>Elucidating how subtle structural differences among positional isomers influence anticancer activity is important for drug development, yet conventional molecular representations and IC50 prediction models do not adequately resolve these differences. Here, we introduce a fixed-reference relative molecular representation and an interpretable ranking framework to quantify how positional-isomer-level structural differences contribute to anticancer activity. Relative relationships between anticancer agents and biomolecules, including lipids and amino acids, were quantified using physicochemical property-related substructures and integrated into cell line-specific models to rank IC50 values within positional isomer groups. Sensitivity analysis revealed distinct structural and physicochemical determinants of anticancer activity across cancer cell lines. We further generated positional isomers of existing anticancer agents and identified candidates predicted to exhibit low IC50 values and tissue-specific activity. These results systematically demonstrate how subtle structural differences among positional isomers affect cell line-specific anticancer activity and provide a strategy for expanding anticancer drug chemical space from known agents.</p>