MOD-PCAC: Multi-Objective Drug Molecule Optimization Algorithm Based on Principal Component Analysis and Clustering

Traditional multi-objective evolutionary algorithms for small molecules suffer from structural homogenization and premature convergence, as they only optimize property space without controlling molecular topology diversity. This work proposes MOD-PCAC, a principal component analysis (PAC) and K-Means clustering based optimization framework. Normalized property vectors and dimension-reduced molecular fingerprints are fused to build a joint feature space that embeds structural information into evolutionary screening. A clustering-guided update strategy is designed: inter-cluster sparsity allocates retention quotas, while dynamic Tanimoto thresholds filter redundant structures; cross-cluster crossover broadens chemical space exploration. Six benchmark datasets and four quantitative metrics are adopted for evaluation. Comparative and ablation experiments verify that MOD-PCAC achieves superior convergence and richer scaffold diversity, effectively improving sampling efficiency for multi-property molecular computational optimization.