Search for: All records

Abstract MotivationEstimation of protein complex structure accuracy is essential for effective structural model selection in structural biology applications such as protein function analysis and drug design. Despite the success of structure prediction methods such as AlphaFold2 and AlphaFold3, selecting top-quality structural models from large model pools remains challenging. ResultsWe present GATE, a novel method that uses graph transformers on pairwise model similarity graphs to predict the quality (accuracy) of complex structural models. By integrating single-model and multimodel quality features, GATE captures intrinsic model characteristics and intermodel geometric similarities to make robust predictions. On the dataset of the 15th Critical Assessment of Protein Structure Prediction (CASP15), GATE achieved the highest Pearson’s correlation (0.748) and the lowest ranking loss (0.1191) compared with existing methods. In the blind CASP16 experiment, GATE ranked fifth based on the sum of z-scores, with a Pearson’s correlation of 0.7076 (first), a Spearman’s correlation of 0.4514 (fourth), a ranking loss of 0.1221 (third), and an area under the curve score of 0.6680 (third) on per-target TM-score-based metrics. Additionally, GATE also performed consistently on large in-house datasets generated by extensive AlphaFold-based sampling with MULTICOM4, confirming its robustness and practical applicability in real-world model selection scenarios. Availability and implementationGATE is available at https://github.com/BioinfoMachineLearning/GATE. 

Motivation. Proteins interact with a variety of molecules, including other proteins, DNAs, RNAs, ligands, ions, and lipids. These interactions play a crucial role in cellular communication, metabolic regulation, gene regulation, and structural integrity, making proteins fundamental to nearly all biological functions. Accurately predicting protein interaction (binding) sites is essential for understanding protein interaction and function. ResultsIn this work, we introduce MPBind, a multitask protein binding site prediction method, which integrates protein language models (PLMs) that can extract structural and functional information from sequences and equivariant graph neural networks (EGNNs) that can effectively capture geometric features of 3D protein structures. Through multitask learning, it can predict binding sites on proteins that interact with five key categories of binding partners: proteins, DNA/RNA, ligands, lipids, and ions. MPBind generalizes across the five molecular classes with state-of-the-art accuracy, achieving AUROC scores of 0.83 and 0.81 for protein–protein and protein–DNA/RNA-binding site prediction, respectively. Moreover, MPBind outperforms both general and task-specific binding site prediction methods, making it a useful, versatile tool for protein binding site prediction. Availability and implementationThe source code of MPBind is available at the GitHub repository: https://github.com/jianlin-cheng/MPBind. 

Abstract The Critical Assessment of Functional Annotation (CAFA) is a long-standing community effort to independently assess computational methods for protein function prediction, to highlight wellperforming methodologies, to identify bottlenecks in the field, and to provide a forum for the dissemination of results and exchange of ideas. In its fifth round (CAFA5) of triennial challenges, a partnership with Kaggle Inc. facilitated participation from a large community of data scientists and computational biologists through a competitive prospective challenge on the crowdsourcing platform. In this work, we present an in-depth analysis of the submitted predictions and report improvements in accuracy over all methods from the previous CAFA challenges. We further introduce a new evaluation setting for proteins with pre-existing (incomplete) annotations and identify the need for methods that better leverage existing annotations to predict those that will be discovered later. Finally, we characterize the prospective evaluation framework by examining performance on a strict set of unpublished annotations and across intermediate database releases. Our results indicate that recent developments in the field, such as the availability of protein language models and accurately predicted 3D structures, as well as the growth of experimental annotations through biocuration, have all contributed to performance improvements. 

ABSTRACT With AlphaFold achieving high‐accuracy tertiary structure prediction for most single‐chain proteins (monomers), the next major challenge in protein structure prediction is to accurately model multichain protein complexes (multimers). We developed MULTICOM4, the latest version of the MULTICOM system, to improve protein complex structure prediction by integrating transformer‐based AlphaFold2, diffusion model‐based AlphaFold3, and our in‐house techniques. These include protein complex stoichiometry prediction, diverse multiple sequence alignment (MSA) generation leveraging both sequence and structure comparison, modeling exception handling, and deep learning‐based protein model quality assessment. MULTICOM4 was blindly evaluated in the 16th Critical Assessment of Techniques for Protein Structure Prediction (CASP16) in 2024. In Phase 0 of CASP16, where stoichiometry information was unavailable, MULTICOM predictors performed best, with MULTICOM_human achieving a TM‐score of 0.752 and a DockQ score of 0.584 for top‐ranked predictions on average. In Phase 1 of CASP16, with stoichiometry information provided, MULTICOM_human remained among the top predictors, attaining a TM‐score of 0.797 and a DockQ score of 0.558 on average. The CASP16 results demonstrate that integrating complementary AlphaFold2 and AlphaFold3 with enhanced MSA inputs, comprehensive model ranking, exception handling, and accurate stoichiometry prediction can effectively improve protein complex structure prediction. 

Abstract The effects of ligand binding on protein structures and their in vivo functions carry numerous implications for modern biomedical research and biotechnology development efforts such as drug discovery. Although several deep learning (DL) methods and benchmarks designed for protein–ligand docking have recently been introduced, so far no previous works have systematically studied the behaviour of the latest docking and structure prediction methods within the broadly applicable context of: (1) using predicted (apo) protein structures for docking (for example, for applicability to new proteins); (2) binding multiple (cofactor) ligands concurrently to a given target protein (for example, for enzyme design); and (3) having no previous knowledge of binding pockets (for example, for generalization to unknown pockets). To enable a deeper understanding of the real-world utility of docking methods, we introduce PoseBench, a comprehensive benchmark for broadly applicable protein–ligand docking. PoseBench enables researchers to rigorously and systematically evaluate DL methods for apo-to-holo protein–ligand docking and protein–ligand structure prediction using both primary ligand and multiligand benchmark datasets, the latter of which we introduce to the DL community. Empirically, using PoseBench, we find that: (1) DL cofolding methods generally outperform comparable conventional and DL docking baseline algorithms, but popular methods such as AlphaFold 3 are still challenged by prediction targets with new protein–ligand binding poses; (2) certain DL cofolding methods are highly sensitive to their input multiple sequence alignments, whereas others are not; and (3) DL methods struggle to strike a balance between structural accuracy and chemical specificity when predicting new or multiligand protein targets. 

ABSTRACT Model quality assessment (MQA) remains a critical component of structural bioinformatics for both structure predictors and experimentalists seeking to use predictions for downstream applications. In CASP16, the Evaluation of Model Accuracy (EMA) category featured both global and local quality estimation for multimeric assemblies (QMODE1 and QMODE2), as well as a novel QMODE3 challenge—requiring predictors to identify the best five models from thousands generated by MassiveFold. This paper presents detailed results from several leading CASP16 EMA methods, highlighting the strengths and limitations of the approaches. 

Abstract MotivationAs fewer than 1% of proteins have protein function information determined experimentally, computationally predicting the function of proteins is critical for obtaining functional information for most proteins and has been a major challenge in protein bioinformatics. Despite the significant progress made in protein function prediction by the community in the last decade, the general accuracy of protein function prediction is still not high, particularly for rare function terms associated with few proteins in the protein function annotation database such as the UniProt. ResultsWe introduce TransFew, a new transformer model, to learn the representations of both protein sequences and function labels [Gene Ontology (GO) terms] to predict the function of proteins. TransFew leverages a large pre-trained protein language model (ESM2-t48) to learn function-relevant representations of proteins from raw protein sequences and uses a biological natural language model (BioBert) and a graph convolutional neural network-based autoencoder to generate semantic representations of GO terms from their textual definition and hierarchical relationships, which are combined together to predict protein function via the cross-attention. Integrating the protein sequence and label representations not only enhances overall function prediction accuracy, but delivers a robust performance of predicting rare function terms with limited annotations by facilitating annotation transfer between GO terms. Availability and implementationhttps://github.com/BioinfoMachineLearning/TransFew. 

Abstract BackgroundThe Critical Assessment of Functional Annotation (CAFA) is a community effort held to understand the field of computational protein function prediction. Every three years, since 2010, the organizers initiate an experiment to collect function predictions on a large set of proteins and then evaluate the performance of predicting methods on a subset of proteins that have accumulated experimental annotations between the submission deadline and the evaluation time. CAFA provides an independent and rigorous assessment of the current state of the art, thus leveling the playing field, highlighting successes, revealing bottlenecks, and offering a forum for the exchange of ideas in protein science. Here, we report the results of the fourth CAFA experiment (CAFA4). ResultsCAFA4 featured the participation of 148 methods from 70 research groups on a total of 46,205 unique proteins over a 5-year annotation accumulation phase, the longest in any CAFA. In a comparison across CAFA2-CAFA4 methods, the prediction of Gene Ontology (GO) terms has clearly improved across all three GO aspects and traditional evaluation settings. While not achieving the first rank, several CAFA2 and CAFA3 methods featured in the top ten methods in many evaluations, suggesting that earlier methods still hold relevance. The performance is weaker in the newly introduced “partial knowledge” evaluation category (proteins with experimental annotations before submission deadline that gained additional annotations in the same GO aspect during the annotation accumulation phase), highlighting the need for a new class of methods. The rankings of the methods were stable over the years in traditional evaluation settings, but less so in the new partial knowledge evaluation. Overall, the field continues to progress with some influx of new participants. Sustained efforts will be necessary to substantially advance it. 

ABSTRACT Predicting the structure of ligands bound to proteins is a foundational problem in modern biotechnology and drug discovery, yet little is known about how to combine the predictions of protein‐ligand structure (poses) produced by the latest deep learning methods to identify the best poses and how to accurately estimate the binding affinity between a protein target and a list of ligand candidates. Further, a blind benchmarking and assessment of protein‐ligand structure and binding affinity prediction is necessary to ensure it generalizes well to new settings. Towards this end, we introduceMULTICOM_ligand, a deep learning‐based protein‐ligand structure and binding affinity prediction ensemble featuring structural consensus ranking for unsupervised pose ranking and a new deep generative flow matching model for joint structure and binding affinity prediction. Notably,MULTICOM_ligand ranked among the top‐5 ligand prediction methods in both protein‐ligand structure prediction and binding affinity prediction in the 16th Critical Assessment of Techniques for Structure Prediction (CASP16), demonstrating its efficacy and utility for real‐world drug discovery efforts. The source code for MULTICOM_ligand is freely available on GitHub. 

Abstract Predicting protein function is a fundamental yet challenging task that requires integrating diverse biological data modalities to capture complex functional relationships. Traditional machine learning methods often rely on single modalities or combine only a limited number (typically two), without aligning them in a unified representation, thereby constraining predictive accuracy. Moreover, most existing approaches are limited to preselected subsets of Gene Ontology (GO) function terms with sufficient annotations, making the prediction of novel function terms a persistent challenge. Here, we present FunBind, a multimodal AI model that jointly learns from protein sequences, textual descriptions, domain annotations, structural features, and GO terms to enhance prediction accuracy and infer previously unseen functions. FunBind operates in two modes: (1) self-supervised pretraining using contrastive learning to align the sequence modality with other heterogeneous modalities in a unified latent space, enabling unsupervised zero-shot function prediction, and (2) supervised fine-tuning that leverages all modalities for comprehensive and accurate function classification. Our results show that FunBind’s zero-shot capabilities allow it to generalize effectively to novel function terms never encountered before, while its joint fine-tuning strategy substantially outperforms single-modality models and current state-of-the-art approaches in prediction accuracy.