فهرست نشریات علمی دسترسی آزاد ایران

Personalized medicine aims to identify effective anticancer therapies tailored to individual patients, a core goal of precision oncology. Despite significant advances, achieving reliable and accurate drug response prediction remains challenging due to the complexity and heterogeneity of pharmacogenomic data. Motivated by the principle that similar cell lines exhibit similar responses to similar drugs, we propose an enhanced matrix factorization framework incorporating a novel dual similarity regularization strategy. The proposed Dual Similarity-Regularized Matrix Factorization (DSRMF) model constrains the latent representations of cell lines and drugs to preserve biological and chemical similarity relationships, ensuring that similar entities occupy proximate positions in the latent space while dissimilar ones remain distant. The model integrates two-dimensional (2D) and three-dimensional (3D) chemical structural features to construct a refined drug similarity matrix and was trained and validated on processed datasets from the Genomics of Drug Sensitivity in Cancer (GDSC) and the Cancer Cell Line Encyclopedia (CCLE). Experimental results demonstrate that DSRMF achieves robust predictive performance, with an average Pearson correlation coefficient (PCC) of approximately 0.96 and a root mean square error (RMSE) of 0.30, indicating a strong correlation between predicted and observed drug responses. These findings confirm that incorporating dual similarity regularization and heterogeneous biological information enhances both predictive accuracy and interpretability. Overall, DSRMF advances drug response modeling and provides a scalable framework for integrating multi-dimensional biological data to improve personalized cancer treatment strategies.