Allostery refers to the biological process by which an effector modulator binds to a protein at a site distant from the active site, known as allosteric site. Identifying allosteric sites is essential for discovering allosteric process and is considered a critical factor in allosteric drug development. To facilitate related research, we developed PASSer (Protein Allosteric Sites Server) at https://passer.smu.edu, a web application for fast and accurate allosteric site prediction and visualization. The website hosts three trained and published machine learning models: (i) an ensemble learning model with extreme gradient boosting and graph convolutional neural network, (ii) an automated machine learning model with AutoGluon and (iii) a learning-to-rank model with LambdaMART. PASSer accepts protein entries directly from the Protein Data Bank (PDB) or user-uploaded PDB files, and can conduct predictions within seconds. The results are presented in an interactive window that displays protein and pockets’ structures, as well as a table that summarizes predictions of the top three pockets with the highest probabilities/scores. To date, PASSer has been visited over 49 000 times in over 70 countries and has executed over 6 200 jobs.

Allostery is considered important in regulating protein’s activity. Drug development depends on the understanding of allosteric mechanisms, especially the identification of allosteric sites, which is a prerequisite in drug discovery and design. Many computational methods have been developed for allosteric site prediction using pocket features and protein dynamics. Here, we present an ensemble learning method, consisting of eXtreme gradient boosting and graph convolutional neural network, to predict allosteric sites. Our model can learn physical properties and topology without any prior information, and shows good performance under multiple indicators. Prediction results showed that 84.9% of allosteric pockets in the test set appeared in the top 3 positions. The PASSer: Protein Allosteric Sites Server (https://passer.smu.edu), along with a command line interface (https://github.com/smutaogroup/passerCLI) provide insights for further analysis in drug discovery.

Allostery is a fundamental process in regulating protein activities. The discovery, design, and development of allosteric drugs demand better identification of allosteric sites. Several computational methods have been developed previously to predict allosteric sites using static pocket features and protein dynamics. Here, we define a baseline model for allosteric site prediction and present a computational model using automated machine learning. Our model, PASSer2.0, advanced the previous results and performed well across multiple indicators with 82.7% of allosteric pockets appearing among the top three positions. The trained machine learning model has been integrated with the Protein Allosteric Sites Server (PASSer) to facilitate allosteric drug discovery.

Allostery plays a crucial role in regulating protein activity, making it a highly sought‐after target in drug development. One of the major challenges in allosteric drug research is the identification of allosteric sites. In recent years, many computational models have been developed for accurate allosteric site prediction. Most of these models focus on designing a general rule that can be applied to pockets of proteins from various families. In this study, we present a new approach using the concept of Learning to Rank (LTR). The LTR model ranks pockets based on their relevance to allosteric sites, that is, how well a pocket meets the characteristics of known allosteric sites. After the training and validation on two datasets, the Allosteric Database (ASD) and CASBench, the LTR model was able to rank an allosteric pocket in the top three positions for 83.6% and 80.5% of test proteins, respectively. The model outperforms other common machine learning models with higher F1 scores (0.662 in ASD and 0.608 in CASBench) and Matthews correlation coefficients (0.645 in ASD and 0.589 in CASBench). The trained model is available on the PASSer platform (https://passer.smu.edu) to aid in drug discovery research.

With the wide usage of organic compounds, the assessment of their acute toxicity has drawn great attention to reduce animal testing and human labor. The development of graph models provides new opportunities for acute toxicity prediction. In this study, five graph models (message-passing neural network, graph convolution network, graph attention network, path-augmented graph transformer network, and Attentive FP) were applied on four toxicity tasks (fish, Daphnia magna, Tetrahymena pyriformis, and Vibrio fischeri). With the lowest prediction error, Attentive FP was reported to have the best performance in all four tasks. Moreover, the attention weights of the Attentive FP model helped to construct atomic heatmaps and provide good explainability.