MCLRP: enhanced prediction of anticancer drug response through low-rank matrix completion and transcriptomic profiling.

BACKGROUND: Accurate prediction of anticancer drug responses remains a significant challenge due to the intricate interplay between genomic features and pharmacological mechanisms. We present Matrix Completion with Low-rank Regularization and Principal Component Analysis (MCLRP), a multimodal framework that synergistically integrates low-rank matrix completion with transcriptomic principal component analysis through dual-stream feature interaction. This innovative architecture not only leverages the similarities among drugs and mutation patterns in cell lines via matrix completion but also preserves gene-level interpretability of response patterns by incorporating gene expression data into the model.

RESULTS: Benchmarked against seven computational paradigms (including matrix completion, ridge regression, SRMF, and their hybrid variants) across the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) repositories, MCLRP demonstrated superior predictive performance for 75% of drug responses, alongside enhanced biological plausibility. Notably, the model identified imatinib as a potential therapeutic alternative for M14 melanoma cell lines through cross-drug response extrapolation, suggesting innovative strategies for overcoming doxorubicin resistance. Interestingly, our mutation-response mapping revealed that BRAF-mutated lineages exhibited a 4.7-fold increase in sensitivity (p < 1e-5) to AZ628 compared to wild-type lineages, with synergistic amplification (8.1-fold, p < 1e-7) observed in BRAF/PIK3CA co-mutants.

CONCLUSIONS: These findings establish MCLRP as a dual-purpose predictive-analytical tool that not only enhances drug response forecasting but also uncovers mutation-specific pharmacological vulnerabilities through systems-level pattern recognition.