PASSerRank: Prediction of allosteric sites with learning to rank

Abstract

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.

Markov State Models and Perturbation-Based Approaches Reveal Distinct Dynamic Signatures and Hidden Allosteric Pockets in the Emerging SARS-Cov-2 Spike Omicron Variant Complexes with the Host Receptor: The Interplay of Dynamics and Convergent Evolution Modulates Allostery and Functional Mechanisms

“…84−87 PASSer offers three pretrained machine learning-based models: (a) an ensemble learning model consisting of extreme gradient boosting (XGBoost) and a graph convolutional neural network (GCNN), 84 (b) an automated machine learning model powered by AutoGluon from Amazon Web Services (AWS), 85 and (c) a learning-torank (LTR) model. 86 The PASSer web server has been widely used for the validation of known functional pockets and the discovery of new allosteric sites. 87 Here, we used the LTR model capable of ranking binding pockets and identifying the most probable allosteric binding sites.…”

“…87 Here, we used the LTR model capable of ranking binding pockets and identifying the most probable allosteric binding sites. 86 The LTR model in PASSer was trained and validated on two widely used data sets, the Allosteric Database (ASD) 126,127 which is a comprehensive database of allosteric proteins and modulators and CASBench which is a benchmarking set that includes annotated catalytic and allosteric sites. 128 FPocket is used in this approach to detect protein binding pockets and calculate physical and chemical features.…”

“…LambdaMART combines gradient boosting decision tree (GBDT) with the loss function derived from LambdaRank, an LTR algorithm. 86 ■ RESULTS AND DISCUSSION Microsecond MD Simulations of the Omicron RBD-ACE2 Complexes Reveal Distinct Dynamic Signatures of RBD Proteins. We first performed a comparative structural analysis of the Omicron RBD-ACE2 complexes for XBB.1 and XBB.1.5 complexes with the ACE2 receptor (Figure 1).…”

Examining Functional Linkages Between Conformational Dynamics, Protein Stability and Evolution of Cryptic Binding Pockets in the SARS-CoV-2 Omicron Spike Complexes with the ACE2 Host Receptor: Recombinant Omicron Variants Mediate Variability of Conserved Allosteric Sites and Binding Epitopes

“…P2Rank v2.4 with default parameters was deployed to identify pockets across all of the representative states from our simulations. By combining eXtreme gradient boosting (XGBoost) and graph convolutional neural networks (GCNNs) a robust approach for allosteric site identification and Prediction of Allosteric Sites Server (PASSer) was developed [98][99][100]. We also employed the PASSer Learning to Rank (LTR) model that is capable of ranking pockets in order of their likelihood to be allosteric sites [100].…”

“…By combining eXtreme gradient boosting (XGBoost) and graph convolutional neural networks (GCNNs) a robust approach for allosteric site identification and Prediction of Allosteric Sites Server (PASSer) was developed [98][99][100]. We also employed the PASSer Learning to Rank (LTR) model that is capable of ranking pockets in order of their likelihood to be allosteric sites [100]. Using P2Rank [96,97] and PASSer LTR [100] approaches, we identified binding pockets in the conformational ensembles and computed P2Rank-predicted residue pocket probability.…”