Anticancer Drug Response Prediction Using Deep Learning

Precision medicine demands accurate prediction of individual patient responses to drugs, a challenge addressed by various computational methods leveraging genomic profiles or chemical structures. This study introduces a novel deep-learning model designed to amalgamate genetic mutation, gene expression, and cheminformatics data within a multi-task convolutional architecture. we identified pertinent cancer-related genes from oncology genetics databases and L1000 landmark genes. The model, incorporating gene expression and mutations as genomic features, and Chemoinformatics features from PubChem or ChEMBL, exhibited enhanced predictive performance through the synergistic integration of these diverse data types. Our approach extends beyond conventional Graph Neural Networks (GNNs) by replacing them with Recurrent Neural Networks (RNNs) in the variants autoencoder deep-learning network. Additionally, we augmented the model's performance by increasing the number of layers in the Convolutional Neural Network (CNN). The extended architecture, complemented by multi-tasking and self-attention functions, demonstrated superior predictive capabilities compared to recently published methods on the same datasets. Our deep learning model has demonstrated improved performance as compared to the previous models. The results underscore the potential of combining gene expression, genetic mutation, and Chemoinformatics features to enhance the accuracy and effectiveness of predictive modeling in the context of precision medicine.