Ultrafast end-to-end protein structure prediction enables high-throughput exploration of uncharacterized proteins

Freeprotmap: waiting-free prediction method for protein distance map

BACKGROUND

Protein residue-residue distance maps are used for remote homology detection, protein information estimation, and protein structure research. However, existing prediction approaches are time-consuming, and hundreds of millions of proteins are discovered each year, necessitating the development of a rapid and reliable prediction method for protein residue-residue distances. Moreover, because many proteins lack known homologous sequences, a waiting-free and alignment-free deep learning method is needed.

RESULT

In this study, we propose a learning framework named FreeProtMap. In terms of protein representation processing, the proposed group pooling in FreeProtMap effectively mitigates issues arising from high-dimensional sparseness in protein representation. In terms of model structure, we have made several careful designs. Firstly, it is designed based on the locality of protein structures and triangular inequality distance constraints to improve prediction accuracy. Secondly, inference speed is improved by using additive attention and lightweight design. Besides, the generalization ability is improved by using bottlenecks and a neural network block named local microformer. As a result, FreeProtMap can predict protein residue-residue distances in tens of milliseconds and has higher precision than the best structure prediction method.

CONCLUSION

Several groups of comparative experiments and ablation experiments verify the effectiveness of the designs. The results demonstrate that FreeProtMap significantly outperforms other state-of-the-art methods in accurate protein residue-residue distance prediction, which is beneficial for lots of protein research works. It is worth mentioning that we could scan all proteins discovered each year based on FreeProtMap to find structurally similar proteins in a short time because the fact that the structure similarity calculation method based on distance maps is much less time-consuming than algorithms based on 3D structures.

Recent Advances and Challenges in Protein Structure Prediction

Artificial intelligence has made significant advances in the field of protein structure prediction in recent years. In particular, DeepMind's end-to-end model, AlphaFold2, has demonstrated the capability to predict three-dimensional structures of numerous unknown proteins with accuracy levels comparable to those of experimental methods. This breakthrough has opened up new possibilities for understanding protein structure and function as well as accelerating drug discovery and other applications in the field of biology and medicine. Despite the remarkable achievements of artificial intelligence in the field, there are still some challenges and limitations. In this Review, we discuss the recent progress and some of the challenges in protein structure prediction. These challenges include predicting multidomain protein structures, protein complex structures, multiple conformational states of proteins, and protein folding pathways. Furthermore, we highlight directions in which further improvements can be conducted.

Guiding discovery of protein sequence-structure-function modeling

MOTIVATION

Protein engineering techniques are key in designing novel catalysts for a wide range of reactions. Although approaches vary in their exploration of the sequence-structure-function paradigm, they are often hampered by the labor-intensive steps of protein expression and screening. In this work, we describe the development and testing of a high-throughput in silico sequence-structure-function pipeline using AlphaFold2 and fast Fourier transform docking that is benchmarked with enantioselectivity and reactivity predictions for an ancestral sequence library of fungal flavin-dependent monooxygenases.

RESULTS

The predicted enantioselectivities and reactivities correlate well with previously described screens of an experimentally available subset of these proteins and capture known changes in enantioselectivity across the phylogenetic tree representing ancestorial proteins from this family. With this pipeline established as our functional screen, we apply ensemble decision tree models and explainable AI techniques to build sequence-function models and extract critical residues within the binding site and the second-sphere residues around this site. We demonstrate that the top-identified key residues in the control of enantioselectivity and reactivity correspond to experimentally verified residues. The in silico sequence-to-function pipeline serves as an accelerated framework to inform protein engineering efforts from vast informative sequence landscapes contained in protein families, ancestral resurrects, and directed evolution campaigns.

AVAILABILITY

Jupyter notebooks detailing the sequence-structure-function pipeline are available at https://github.com/BrooksResearchGroup-UM/seq_struct_func.

Repurposing dye ligands as antivirals via a docking approach on viral membrane and globular proteins - SARS-CoV-2 and HPV-16

A series of dye ligands are docked to three different proteins, E and 3a of severe acute respiratory syndrome corona virus 2 (SARS-CoV-2) and E6 of human papilloma virus type 16 (HPV-16) using three different software. A four-level selection algorithm is used based on nonparametric statistics of numerical key values such as the "rank" derived from (i) averaged estimated binding energies (EBEs) and (ii) absolute EBE value of each of the software, (iii) frequency of ranking and (iv) rank of the area-under-curve values (AUCs) from decoy docking. A series of repurposing drugs and known antivirals used in experimental studies are docked for comparison. One dye ligand is ranked best for all proteins using the selection algorithm levels i - iii. Another three dye ligands are ranked top for the proteins individually when using all four levels.

AlphaFold2 Can Predict Single-Mutation Effects

AlphaFold2 (AF) is a promising tool, but is it accurate enough to predict single mutation effects? Here, we report that the localized structural deformation between protein pairs differing by only 1-3 mutations-as measured by the effective strain-is correlated across 3901 experimental and AF-predicted structures. Furthermore, analysis of ∼11 000 proteins shows that the local structural change correlates with various phenotypic changes. These findings suggest that AF can predict the range and magnitude of single-mutation effects on average, and we propose a method to improve precision of AF predictions and to indicate when predictions are unreliable.

Machine Learning-Guided Protein Engineering

Recent progress in engineering highly promising biocatalysts has increasingly involved machine learning methods. These methods leverage existing experimental and simulation data to aid in the discovery and annotation of promising enzymes, as well as in suggesting beneficial mutations for improving known targets. The field of machine learning for protein engineering is gathering steam, driven by recent success stories and notable progress in other areas. It already encompasses ambitious tasks such as understanding and predicting protein structure and function, catalytic efficiency, enantioselectivity, protein dynamics, stability, solubility, aggregation, and more. Nonetheless, the field is still evolving, with many challenges to overcome and questions to address. In this Perspective, we provide an overview of ongoing trends in this domain, highlight recent case studies, and examine the current limitations of machine learning-based methods. We emphasize the crucial importance of thorough experimental validation of emerging models before their use for rational protein design. We present our opinions on the fundamental problems and outline the potential directions for future research.

When will RNA get its AlphaFold moment?

The protein structure prediction problem has been solved for many types of proteins by AlphaFold. Recently, there has been considerable excitement to build off the success of AlphaFold and predict the 3D structures of RNAs. RNA prediction methods use a variety of techniques, from physics-based to machine learning approaches. We believe that there are challenges preventing the successful development of deep learning-based methods like AlphaFold for RNA in the short term. Broadly speaking, the challenges are the limited number of structures and alignments making data-hungry deep learning methods unlikely to succeed. Additionally, there are several issues with the existing structure and sequence data, as they are often of insufficient quality, highly biased and missing key information. Here, we discuss these challenges in detail and suggest some steps to remedy the situation. We believe that it is possible to create an accurate RNA structure prediction method, but it will require solving several data quality and volume issues, usage of data beyond simple sequence alignments, or the development of new less data-hungry machine learning methods.

Protein Structure Prediction: Challenges, Advances, and the Shift of Research Paradigms

Protein structure prediction is an interdisciplinary research topic that has attracted researchers from multiple fields, including biochemistry, medicine, physics, mathematics, and computer science. These researchers adopt various research paradigms to attack the same structure prediction problem: biochemists and physicists attempt to reveal the principles governing protein folding; mathematicians, especially statisticians, usually start from assuming a probability distribution of protein structures given a target sequence and then find the most likely structure, while computer scientists formulate protein structure prediction as an optimization problem - finding the structural conformation with the lowest energy or minimizing the difference between predicted structure and native structure. These research paradigms fall into the two statistical modeling cultures proposed by Leo Breiman, namely, data modeling and algorithmic modeling. Recently, we have also witnessed the great success of deep learning in protein structure prediction. In this review, we present a survey of the efforts for protein structure prediction. We compare the research paradigms adopted by researchers from different fields, with an emphasis on the shift of research paradigms in the era of deep learning. In short, the algorithmic modeling techniques, especially deep neural networks, have considerably improved the accuracy of protein structure prediction; however, theories interpreting the neural networks and knowledge on protein folding are still highly desired.

Open-Source Machine Learning in Computational Chemistry

The field of computational chemistry has seen a significant increase in the integration of machine learning concepts and algorithms. In this Perspective, we surveyed 179 open-source software projects, with corresponding peer-reviewed papers published within the last 5 years, to better understand the topics within the field being investigated by machine learning approaches. For each project, we provide a short description, the link to the code, the accompanying license type, and whether the training data and resulting models are made publicly available. Based on those deposited in GitHub repositories, the most popular employed Python libraries are identified. We hope that this survey will serve as a resource to learn about machine learning or specific architectures thereof by identifying accessible codes with accompanying papers on a topic basis. To this end, we also include computational chemistry open-source software for generating training data and fundamental Python libraries for machine learning. Based on our observations and considering the three pillars of collaborative machine learning work, open data, open source (code), and open models, we provide some suggestions to the community.

Improved structure-related prediction for insufficient homologous proteins using MSA enhancement and pre-trained language model

In recent years, protein structure problems have become a hotspot for understanding protein folding and function mechanisms. It has been observed that most of the protein structure works rely on and benefit from co-evolutionary information obtained by multiple sequence alignment (MSA). As an example, AlphaFold2 (AF2) is a typical MSA-based protein structure tool which is famous for its high accuracy. As a consequence, these MSA-based methods are limited by the quality of the MSAs. Especially for orphan proteins that have no homologous sequence, AlphaFold2 performs unsatisfactorily as MSA depth decreases, which may pose a barrier to its widespread application in protein mutation and design problems in which there are no rich homologous sequences and rapid prediction is needed. In this paper, we constructed two standard datasets for orphan and de novo proteins which have insufficient/none homology information, called Orphan62 and Design204, respectively, to fairly evaluate the performance of the various methods in this case. Then, depending on whether or not utilizing scarce MSA information, we summarized two approaches, MSA-enhanced and MSA-free methods, to effectively solve the issue without sufficient MSAs. MSA-enhanced model aims to improve poor MSA quality from the data source by knowledge distillation and generation models. MSA-free model directly learns the relationship between residues on enormous protein sequences from pre-trained models, bypassing the step of extracting the residue pair representation from MSA. Next, we evaluated the performance of four MSA-free methods (trRosettaX-Single, TRFold, ESMFold and ProtT5) and MSA-enhanced (Bagging MSA) method compared with a traditional MSA-based method AlphaFold2, in two protein structure-related prediction tasks, respectively. Comparison analyses show that trRosettaX-Single and ESMFold which belong to MSA-free method can achieve fast prediction ($\sim\! 40$s) and comparable performance compared with AF2 in tertiary structure prediction, especially for short peptides, $\alpha $-helical segments and targets with few homologous sequences. Bagging MSA utilizing MSA enhancement improves the accuracy of our trained base model which is an MSA-based method when poor homology information exists in secondary structure prediction. Our study provides biologists an insight of how to select rapid and appropriate prediction tools for enzyme engineering and peptide drug development.

CONTACT

guofei@csu.edu.cn, jj.tang@siat.ac.cn.

Machine Learning Advances in Predicting Peptide/Protein-Protein Interactions Based on Sequence Information for Lead Peptides Discovery

Peptides have shown increasing advantages and significant clinical value in drug discovery and development. With the development of high-throughput technologies and artificial intelligence (AI), machine learning (ML) methods for discovering new lead peptides have been expanded and incorporated into rational drug design. Predictions of peptide-protein interactions (PepPIs) and protein-protein interactions (PPIs) are both opportunities and challenges in computational biology, which will help to better understand the mechanisms of disease and provide the impetus for the discovery of lead peptides. This paper comprehensively reviews computational models for PepPI and PPI predictions. It begins with an introduction of various databases of peptide ligands and target proteins. Then it discusses data formats and feature representations for proteins and peptides. Furthermore, classical ML methods and emerging deep learning (DL) methods that can be used to train prediction models of PepPI and PPI are classified into four categories, and their advantages and disadvantages are analyzed. To assess the relative performance of different models, different validation protocols and evaluation indexes are discussed. The goal of this review is to help researchers quickly get started to develop computational frameworks using these integrated resources and eventually promote the discovery of lead peptides.

Progress at protein structure prediction, as seen in CASP15

In Dec 2020, the results of AlphaFold version 2 were presented at CASP14, sparking a revolution in the field of protein structure predictions. For the first time, a purely computational method could challenge experimental accuracy for structure prediction of single protein domains. The code of AlphaFold v2 was released in the summer of 2021, and since then, it has been shown that it can be used to accurately predict the structure of most ordered proteins and many protein-protein interactions. It has also sparked an explosion of development in the field, improving AI-based methods to predict protein complexes, disordered regions, and protein design. Here I will review some of the inventions sparked by the release of AlphaFold.

CoaDTI: multi-modal co-attention based framework for drug-target interaction annotation

MOTIVATION

The identification of drug-target interactions (DTIs) plays a vital role for in silico drug discovery, in which the drug is the chemical molecule, and the target is the protein residues in the binding pocket. Manual DTI annotation approaches remain reliable; however, it is notoriously laborious and time-consuming to test each drug-target pair exhaustively. Recently, the rapid growth of labelled DTI data has catalysed interests in high-throughput DTI prediction. Unfortunately, those methods highly rely on the manual features denoted by human, leading to errors.

RESULTS

Here, we developed an end-to-end deep learning framework called CoaDTI to significantly improve the efficiency and interpretability of drug target annotation. CoaDTI incorporates the Co-attention mechanism to model the interaction information from the drug modality and protein modality. In particular, CoaDTI incorporates transformer to learn the protein representations from raw amino acid sequences, and GraphSage to extract the molecule graph features from SMILES. Furthermore, we proposed to employ the transfer learning strategy to encode protein features by pre-trained transformer to address the issue of scarce labelled data. The experimental results demonstrate that CoaDTI achieves competitive performance on three public datasets compared with state-of-the-art models. In addition, the transfer learning strategy further boosts the performance to an unprecedented level. The extended study reveals that CoaDTI can identify novel DTIs such as reactions between candidate drugs and severe acute respiratory syndrome coronavirus 2-associated proteins. The visualization of co-attention scores can illustrate the interpretability of our model for mechanistic insights.

AVAILABILITY

Source code are publicly available at https://github.com/Layne-Huang/CoaDTI.

Prediction of inter-chain distance maps of protein complexes with 2D attention-based deep neural networks

Residue-residue distance information is useful for predicting tertiary structures of protein monomers or quaternary structures of protein complexes. Many deep learning methods have been developed to predict intra-chain residue-residue distances of monomers accurately, but few methods can accurately predict inter-chain residue-residue distances of complexes. We develop a deep learning method CDPred (i.e., Complex Distance Prediction) based on the 2D attention-powered residual network to address the gap. Tested on two homodimer datasets, CDPred achieves the precision of 60.94% and 42.93% for top L/5 inter-chain contact predictions (L: length of the monomer in homodimer), respectively, substantially higher than DeepHomo's 37.40% and 23.08% and GLINTER's 48.09% and 36.74%. Tested on the two heterodimer datasets, the top Ls/5 inter-chain contact prediction precision (Ls: length of the shorter monomer in heterodimer) of CDPred is 47.59% and 22.87% respectively, surpassing GLINTER's 23.24% and 13.49%. Moreover, the prediction of CDPred is complementary with that of AlphaFold2-multimer.

Single-sequence protein structure prediction using a language model and deep learning

AlphaFold2 and related computational systems predict protein structure using deep learning and co-evolutionary relationships encoded in multiple sequence alignments (MSAs). Despite high prediction accuracy achieved by these systems, challenges remain in (1) prediction of orphan and rapidly evolving proteins for which an MSA cannot be generated; (2) rapid exploration of designed structures; and (3) understanding the rules governing spontaneous polypeptide folding in solution. Here we report development of an end-to-end differentiable recurrent geometric network (RGN) that uses a protein language model (AminoBERT) to learn latent structural information from unaligned proteins. A linked geometric module compactly represents Cα backbone geometry in a translationally and rotationally invariant way. On average, RGN2 outperforms AlphaFold2 and RoseTTAFold on orphan proteins and classes of designed proteins while achieving up to a 106-fold reduction in compute time. These findings demonstrate the practical and theoretical strengths of protein language models relative to MSAs in structure prediction.

E-SNPs&GO: embedding of protein sequence and function improves the annotation of human pathogenic variants

MOTIVATION

The advent of massive DNA sequencing technologies is producing a huge number of human single-nucleotide polymorphisms occurring in protein-coding regions and possibly changing their sequences. Discriminating harmful protein variations from neutral ones is one of the crucial challenges in precision medicine. Computational tools based on artificial intelligence provide models for protein sequence encoding, bypassing database searches for evolutionary information. We leverage the new encoding schemes for an efficient annotation of protein variants.

RESULTS

E-SNPs&GO is a novel method that, given an input protein sequence and a single amino acid variation, can predict whether the variation is related to diseases or not. The proposed method adopts an input encoding completely based on protein language models and embedding techniques, specifically devised to encode protein sequences and GO functional annotations. We trained our model on a newly generated dataset of 101 146 human protein single amino acid variants in 13 661 proteins, derived from public resources. When tested on a blind set comprising 10 266 variants, our method well compares to recent approaches released in literature for the same task, reaching a Matthews Correlation Coefficient score of 0.72. We propose E-SNPs&GO as a suitable, efficient and accurate large-scale annotator of protein variant datasets.

AVAILABILITY AND IMPLEMENTATION

The method is available as a webserver at https://esnpsandgo.biocomp.unibo.it. Datasets and predictions are available at https://esnpsandgo.biocomp.unibo.it/datasets.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

AlphaFold2 models indicate that protein sequence determines both structure and dynamics

AlphaFold 2 (AF2) has placed Molecular Biology in a new era where we can visualize, analyze and interpret the structures and functions of all proteins solely from their primary sequences. We performed AF2 structure predictions for various protein systems, including globular proteins, a multi-domain protein, an intrinsically disordered protein (IDP), a randomized protein, two larger proteins (> 1000 AA), a heterodimer and a homodimer protein complex. Our results show that along with the three dimensional (3D) structures, AF2 also decodes protein sequences into residue flexibilities via both the predicted local distance difference test (pLDDT) scores of the models, and the predicted aligned error (PAE) maps. We show that PAE maps from AF2 are correlated with the distance variation (DV) matrices from molecular dynamics (MD) simulations, which reveals that the PAE maps can predict the dynamical nature of protein residues. Here, we introduce the AF2-scores, which are simply derived from pLDDT scores and are in the range of [0, 1]. We found that for most protein models, including large proteins and protein complexes, the AF2-scores are highly correlated with the root mean square fluctuations (RMSF) calculated from MD simulations. However, for an IDP and a randomized protein, the AF2-scores do not correlate with the RMSF from MD, especially for the IDP. Our results indicate that the protein structures predicted by AF2 also convey information of the residue flexibility, i.e., protein dynamics.

Improved prediction of protein-protein interactions using AlphaFold2

Predicting the structure of interacting protein chains is a fundamental step towards understanding protein function. Unfortunately, no computational method can produce accurate structures of protein complexes. AlphaFold2, has shown unprecedented levels of accuracy in modelling single chain protein structures. Here, we apply AlphaFold2 for the prediction of heterodimeric protein complexes. We find that the AlphaFold2 protocol together with optimised multiple sequence alignments, generate models with acceptable quality (DockQ ≥ 0.23) for 63% of the dimers. From the predicted interfaces we create a simple function to predict the DockQ score which distinguishes acceptable from incorrect models as well as interacting from non-interacting proteins with state-of-art accuracy. We find that, using the predicted DockQ scores, we can identify 51% of all interacting pairs at 1% FPR.

Predicting the structure of protein complexes is extremely difficult. Here, authors apply AlphaFold2 with optimized multiple sequence alignments to model complexes of interacting proteins, enabling prediction of both if and how proteins interact with state-of-art accuracy.