Background: Accurately predicting drug-target binding affinity (DTA) in silico plays an important role in drug discovery. Most of the computational methods developed for predicting DTA use machine learning models, especially deep neural networks, and depend on large-scale labelled data. However, it is difficult to learn enough feature representation from tens of millions of compounds and hundreds of thousands of proteins only based on relatively limited labelled drug-target data. There are a large number of unknown drugs, which never appear in the labelled drug-target data. This is a kind of out-of-distribution problems in bio-medicine. Some recent studies adopted self-supervised pre-training tasks to learn structural information of amino acid sequences for enhancing the feature representation of proteins. However, the task gap between pre-training and DTA prediction brings the catastrophic forgetting problem, which hinders the full application of feature representation in DTA prediction and seriously affects the generalization capability of models for unknown drug discovery. Results: To address these problems, we propose the GeneralizedDTA, which is a new DTA prediction model oriented to unknown drug discovery, by combining pre-training and multi-task learning. We introduce self-supervised protein and drug pre-training tasks to learn richer structural information from amino acid sequences of proteins and molecular graphs of drug compounds, in order to alleviate the problem of high variance caused by encoding based on deep neural networks and accelerate the convergence of prediction model on small-scale labelled data. We also develop a multi-task learning framework with a dual adaptation mechanism to narrow the task gap between pre-training and prediction for preventing overfitting and improving the generalization capability of DTA prediction model on unknown drug discovery. To validate the effectiveness of our model, we construct an unknown drug data set to simulate the scenario of unknown drug discovery. Compared with existing DTA prediction models, the experimental results show that our model has the higher generalization capability in the DTA prediction of unknown drugs. Conclusions: The advantages of our model are mainly attributed to two kinds of pre-training tasks and the multi-task learning framework, which can learn richer structural information of proteins and drugs from large-scale unlabeled data, and then effectively integrate it into the downstream prediction task for obtaining a high-quality DTA prediction in unknown drug discovery.

Depression is a complex mental disease, which involves many factors such as psychology, physiology, and society, and which causes harm to society. Up to now, there are no valuable biomarkers for clinical diagnosis. This research constructed a dataset, which includes calm, sad, and happy facial expressions from both patients with depression and normal people, and classification and visualization of depression. The network includes a dual-scale convolution module, adaptive channel attentional mechanism, and gradient class activation mapping technique. In which, dual-scale convolution captures features of the facial region at different scales and the adaptive channel attention captures the facial region with the most significant features. The results show that we improve the performance of depression classification based on facial information, and recruit gradient class activation mapping technique obtaining a specific visual face pattern of depression that is different from that of normal people, which provides a potential interpretable and discriminant evidence for the clinical diagnosis. Thereby, promoting the development and application of artificial intelligence in the field of psychiatry.

Machine learning (ML) techniques have made enormous progress in the field of materials science. However, many conventional ML algorithms operate as “black‐boxes”, lacking transparency in revealing explicit relationships between material features and target properties. To address this, the development of interpretable ML models is essential to drive further advancements in AI‐driven materials discovery. In this study, we present an interpretable framework that combines traditional machine learning with symbolic regression, using Janus III–VI vdW heterostructures as a case study. This approach enables fast and accurate predictions of stability and electronic structure. Our results demonstrate that the prediction accuracy using the classification model for stability, based on formation energy, reaches 0.960. On the other hand, the R², MAE, and RMSE value using the regression model for electronic structure prediction, based on band gap, achieves 0.927, 0.113, and 0.141 on the testing set, respectively. Additionally, we identify a universal interpretable descriptor comprising five simple parameters that reveals the underlying physical relationships between the candidate heterostructures and their band gaps. This descriptor not only delivers high accuracy in band gap prediction but also provides explicit physical insight into the material properties.

Satisfiability modulo theories (SMT) solvers have been widely applied as the reasoning engine for diverse software analysis and verification technologies. The efficiency of the SMT solver has significant effects on the performance of these technologies. However, the current SMT solvers are designed for the general purpose of constraint solving. Many useful knowledge of programs cannot be utilized during the SMT solving. As a result, the SMT solver may spend a lot of effort to explore redundant search space. In this paper, we propose a novel approach for utilizing control-flow knowledge in SMT solving. With this technique, the search space can be considerably reduced and the efficiency of SMT solving is observably improved. We conducted extensive experiments on credible benchmarks, the results show orders of magnitude improvements of our approach.