DBDNMF: A Dual Branch Deep Neural Matrix Factorization method for drug response prediction

Area of Science:

• Computational biology
• Bioinformatics
• Pharmacogenomics

Background:

• Individualized anti-cancer drug response prediction is vital for precision medicine.
• Wet-lab experiments for drug response prediction are costly and time-consuming.
• Computational models can offer efficient alternatives for predicting drug-cell line interactions.

Purpose of the Study:

• To develop a computational model for precise prediction of anti-cancer drug responses.
• To address limitations of existing methods that focus on either linear or nonlinear relationships.
• To improve decision-making in precision medicine through accurate drug response prediction.

Main Methods:

• Proposed a Dual Branch Deep Neural Matrix Factorization (DBDNMF) method.
• DBDNMF learns latent representations of drugs and cell lines using flexible inputs.
• Reconstructs partially observed drug-response matrices via deep neural network layers.

Main Results:

• DBDNMF demonstrated superior accuracy in predicting drug responses compared to state-of-the-art algorithms on CCLE and GDSC datasets.
• The model proved reliable and stable in its predictions.
• Hierarchical clustering revealed that drugs with similar responses target similar pathways, and cell lines from the same tissue share response patterns.

Conclusions:

• DBDNMF offers a robust and accurate approach for predicting anti-cancer drug responses.
• The findings support the utility of computational models in advancing precision medicine.
• The model's ability to identify drug-pathway and cell line-tissue relationships provides valuable biological insights.