DeepTTA: a transformer-based model for predicting cancer drug response

Abstract

Identifying new lead molecules to treat cancer requires more than a decade of dedicated effort. Before selected drug candidates are used in the clinic, their anti-cancer activity is generally validated by in vitro cellular experiments. Therefore, accurate prediction of cancer drug response is a critical and challenging task for anti-cancer drugs design and precision medicine. With the development of pharmacogenomics, the combination of efficient drug feature extraction methods and omics data has made it possible to use computational models to assist in drug response prediction. In this study, we propose DeepTTA, a novel end-to-end deep learning model that utilizes transformer for drug representation learning and a multilayer neural network for transcriptomic data prediction of the anti-cancer drug responses. Specifically, DeepTTA uses transcriptomic gene expression data and chemical substructures of drugs for drug response prediction. Compared to existing methods, DeepTTA achieved higher performance in terms of root mean square error, Pearson correlation coefficient and Spearman’s rank correlation coefficient on multiple test sets. Moreover, we discovered that anti-cancer drugs bortezomib and dactinomycin provide a potential therapeutic option with multiple clinical indications. With its excellent performance, DeepTTA is expected to be an effective method in cancer drug design.

Deep learning methods for drug response prediction in cancer: Predominant and emerging trends

“…Transformer is an attention-based architecture that have recently been explored in DRP models for encoding drug representations ( 70 , 104 ). In GraTransDRP ( 104 ), the authors propose to extend an existing model, GraOmicDRP ( 105 ), by modifying the drug subnetwork to include graph attention network (GAT), graph isomorphism network (GIN), and a graph transformer which learns from graph data.…”

“…In GraTransDRP ( 104 ), the authors propose to extend an existing model, GraOmicDRP ( 105 ), by modifying the drug subnetwork to include graph attention network (GAT), graph isomorphism network (GIN), and a graph transformer which learns from graph data. In DeepTTA ( 70 ), a transformer module encodes drug information represented as text data [ESPF substructures ( 112 )]. Both models report significant improvement in generalization thanks to transformer modules.…”

“…The second approach is to train regression models to predict one of the continuous responses and then categorize it into two or more classes (green in Figure 4 ). Only a few papers explored this approach, including binary ( 67 – 70 ) and multi-class labels ( 71 ). This approach naturally allows assessing model performance using both regression and classification metrics, possibly offering a more robust generalization analysis.…”

Data augmentation and multimodal learning for predicting drug response in patient-derived xenografts from gene expressions and histology images

“…As compared to DL models that are trained with cell line data, all models in Table 2 generally exhibit a relatively lower performance. For example, the average AUPRC is around 0.27 in Table 2 (all precision-recall curves are provided in the Supplementary material ) but models trained on cell lines can exhibit AUPRC of 0.7 and above ( 48 ). Yet, we can observe a large spread of scores for all models and metrics ( Figure 6 ).…”

Machine learning models based on immunological genes to predict the response to neoadjuvant therapy in breast cancer patients

“…For example, an intelligent VAB (5), which is a machine learning algorithm based on clinicopathological, image, and biopsy features, can identify BC patients who have achieved pCR after NAT, and exempt them from surgery. For drug response prediction, many artificial intelligence models have been reported, such as Deep Drug Response (6), DeepDR (7), tCNNS (8), DeepTTA (9), and VAE model (10). Although some complex and state-of-art algorithms were employed in these models, however, they were trained on an in vitro cell line or a pan-cancer dataset.…”