Tumor Drug Response Prediction Based on Attention Network and Transfer Learning Methods
Year: 2025
Type: article
Abstract: Introduction/Objective:: Accurately predicting tumor cell line responses to therapeutic drugs is crucial for personalized cancer treatment. Existing methods often rely on either bulk or single- cell data, limiting their ability to fully capture tumor heterogeneity and the complex mechanisms of drug response. This study aims to develop a novel approach that combines bulk and single-cell data to enhance the accuracy of tumor drug response predictions, focusing on the integration of attention mechanisms and transfer learning. Methods: This paper introduces SCATDP (Single Cell Attention Transfer Learning for Drug Response Prediction). This model leverages an attention mechanism for predicting drug responses in single-cell RNA sequencing (scRNA-seq) data. The model is pre-trained on bulk cell gene expression data and then transfers this knowledge to single-cell data, utilizing a multi-head attention mechanism to focus on gene expression patterns related to drug response. The performance was evaluated using four scRNA-seq datasets, with metrics such as Recall, ROC, and AP used for comparison. Results: SCATDP demonstrated superior performance in predicting drug responses compared to existing methods. It outperformed other models in several key metrics, including Recall, ROC, and AP, highlighting its robust predictive capabilities. The model also provided an improved interpretation of gene expression patterns associated with drug response, showcasing its potential to handle complex tumor environments. Conclusion: SCATDP presents a powerful tool for predicting drug responses in cancer, combining bulk and single-cell data with attention mechanisms to enhance accuracy. This approach offers a solid foundation for personalized cancer treatment decisions and has the potential to improve drug screening and treatment optimization in oncology.
Source: Current Bioinformatics
Authors: BingWei Zhou, SiLin Sun, YueYi Jiang, Qi Dai, Xia Yu, HaiXia Long
Institutions: Hainan Normal University, Zhejiang Sci-Tech University
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4409170252