Transfer learning for drug–target interaction prediction

Abstract

Motivation Utilizing AI-driven approaches for drug–target interaction (DTI) prediction require large volumes of training data which are not available for the majority of target proteins. In this study, we investigate the use of deep transfer learning for the prediction of interactions between drug candidate compounds and understudied target proteins with scarce training data. The idea here is to first train a deep neural network classifier with a generalized source training dataset of large size and then to reuse this pre-trained neural network as an initial configuration for re-training/fine-tuning purposes with a small-sized specialized target training dataset. To explore this idea, we selected six protein families that have critical importance in biomedicine: kinases, G-protein-coupled receptors (GPCRs), ion channels, nuclear receptors, proteases, and transporters. In two independent experiments, the protein families of transporters and nuclear receptors were individually set as the target datasets, while the remaining five families were used as the source datasets. Several size-based target family training datasets were formed in a controlled manner to assess the benefit provided by the transfer learning approach. Results Here, we present a systematic evaluation of our approach by pre-training a feed-forward neural network with source training datasets and applying different modes of transfer learning from the pre-trained source network to a target dataset. The performance of deep transfer learning is evaluated and compared with that of training the same deep neural network from scratch. We found that when the training dataset contains fewer than 100 compounds, transfer learning outperforms the conventional strategy of training the system from scratch, suggesting that transfer learning is advantageous for predicting binders to under-studied targets. Availability and implementation The source code and datasets are available at https://github.com/cansyl/TransferLearning4DTI. Our web-based service containing the ready-to-use pre-trained models is accessible at https://tl4dti.kansil.org.

Dual-view jointly learning improves personalized drug synergy prediction

“…Although the above methods have achieved satisfactory performance under crossvalidation with randomly splitting, but their accuracy decreased significantly for unseen drugs and cancer samples, especially for cross-dataset prediction [29]. With the development of various transfer learning technologies in computational fields [32,33],…”

“…Although the above methods have achieved satisfactory performance under cross-validation with randomly splitting, but their accuracy decreased significantly for unseen drugs and cancer samples, especially for cross-dataset prediction[29]. With the development of various transfer learning technologies in computational fields[32,33], we supposed that fine-tuning may improve the accuracy of drug synergy prediction on new drugs and cancer samples even for a small experimental dataset. Additionally, current methods modeled two drugs independently, did not sufficiently utilize the association between two drugs.…”

Quantifying the Hardness of Bioactivity Prediction Tasks for Transfer Learning

“…Numerous knowledge-sharing methods have been devised and are increasingly used to address the challenges of the persistent data shortage . These include transfer learning, , multitask learning, , and meta-learning. − Common to all knowledge-sharing methods is that they use information from related tasks (“source tasks” or “training tasks”) or data sets to improve the performance on tasks for which typically low and scarce amounts of data are available (“target tasks” or “test tasks”).…”

Data mining of PubChem bioassay records reveals diverse OXPHOS inhibitory chemotypes as potential therapeutic agents against ovarian cancer

“…The key limitation of such models, however, is that they may only be applied to targets for which ample molecule screening data are available for model training. For novel targets where such data are lacking, HTS assay data involving related protein targets might be used for training in a transfer learning-styled approach [5][6][7] . However, in such cases, it may not be clear which screening data sets should be included or whether data from a collection of varied assays can be effectively integrated for SAR-based data mining and model training.…”