Cancer drug response prediction is crucial for precision medicine, as it can improve treatment outcomes and reduce medical costs. However, existing models often ignore the geometric features of drug molecules and their interactions with cancer cells. To address this, this study proposes a multiomics fusion model named MTEGDRP. The model uses a transformer to extract high-level features from drug and cell data, as well as their interactions, while an equivariant graph neural network captures the spatial structure of drugs. In regression tasks, MTEGDRP performs better than current state-of-the-art methods. Ablation studies show that multiomics integration and molecular spatial information are effective. Visualization of the feature weights provides interpretability for the model. With its excellent prediction performance, MTEGDRP shows great potential as a useful tool for guiding anticancer drug design in precision medicine.
Liu et al. (Wed,) studied this question.