Deep Learning and big data have shown tremendous success in bioinformatics and computational biology in recent years; artificial intelligence methods have also significantly contributed in the task of protein function classification. This review paper analyzes the recent developments in approaches for the task of predicting protein function using deep learning. We explain the importance of determining the protein function and why automating the following task is crucial. Then, after reviewing the widely used deep learning techniques for this task, we continue our review and highlight the emergence of the modern State of The Art (SOTA) deep learning models which have achieved groundbreaking results in the field of computer vision, natural language processing and multi-modal learning in the last few years. We hope that this review will provide a broad view of the current role and advances of deep learning in biological sciences, especially in predicting protein function tasks and encourage new researchers to contribute to this area.
Information overload is a major obstacle to scientific progress. The explosive growth in scientific literature and data has made it ever harder to discover useful insights in a large mass of information. Today scientific knowledge is accessed through search engines, but they are unable to organize scientific knowledge alone. In this paper we introduce Galactica: a large language model that can store, combine and reason about scientific knowledge. We train on a large scientific corpus of papers, reference material, knowledge bases and many other sources. We outperform existing models on a range of scientific tasks. On technical knowledge probes such as LaTeX equations, Galactica outperforms the latest GPT-3 by 68.2% versus 49.0%. Galactica also performs well on reasoning, outperforming Chinchilla on mathematical MMLU by 41.3% to 35.7%, and PaLM 540B on MATH with a score of 20.4% versus 8.8%. It also sets a new state-of-the-art on downstream tasks such as PubMedQA and MedMCQA dev of 77.6% and 52.9%. And despite not being trained on a general corpus, Galactica outperforms BLOOM and OPT-175B on BIG-bench. We believe these results demonstrate the potential for language models as a new interface for science. We open source the model for the benefit of the scientific community.
We are now witnessing significant progress of deep learning methods in a variety of tasks (or datasets) of proteins. However, there is a lack of a standard benchmark to evaluate the performance of different methods, which hinders the progress of deep learning in this field. In this paper, we propose such a benchmark called PEER, a comprehensive and multi-task benchmark for Protein sEquence undERstanding. PEER provides a set of diverse protein understanding tasks including protein function prediction, protein localization prediction, protein structure prediction, protein-protein interaction prediction, and protein-ligand interaction prediction. We evaluate different types of sequence-based methods for each task including traditional feature engineering approaches, different sequence encoding methods as well as large-scale pre-trained protein language models. In addition, we also investigate the performance of these methods under the multi-task learning setting. Experimental results show that large-scale pre-trained protein language models achieve the best performance for most individual tasks, and jointly training multiple tasks further boosts the performance. The datasets and source codes of this benchmark are all available at https://github.com/DeepGraphLearning/PEER_Benchmark
Learning from 3D protein structures has gained wide interest in protein modeling and structural bioinformatics. Unfortunately, the number of available structures is orders of magnitude lower than the training data sizes commonly used in computer vision and machine learning. Moreover, this number is reduced even further, when only annotated protein structures can be considered, making the training of existing models difficult and prone to over-fitting. To address this challenge, we introduce a new representation learning framework for 3D protein structures. Our framework uses unsupervised contrastive learning to learn meaningful representations of protein structures, making use of proteins from the Protein Data Bank. We show, how these representations can be used to solve a large variety of tasks, such as protein function prediction, protein fold classification, structural similarity prediction, and protein-ligand binding affinity prediction. Moreover, we show how fine-tuned networks, pre-trained with our algorithm, lead to significantly improved task performance, achieving new state-of-the-art results in many tasks.
Accurately finding proteins and genes that have a certain function is the prerequisite for a broad range of biomedical applications. Despite the encouraging progress of existing computational approaches in protein function prediction, it remains challenging to annotate proteins to a novel function that is not collected in the Gene Ontology and does not have any annotated proteins. This limitation, a side effect from the widely-used multi-label classification problem setting of protein function prediction, hampers the progress of studying new pathways and biological processes, and further slows down research in various biomedical areas. Here, we tackle this problem by annotating proteins to a function only based on its textual description so that we do not need to know any associated proteins for this function. The key idea of our method ProTranslator is to redefine protein function prediction as a machine translation problem, which translates the description word sequence of a function to the amino acid sequence of a protein. We can then transfer annotations from functions that have similar textual description to annotate a novel function. We observed substantial improvement in annotating novel functions and sparsely annotated functions on CAFA3, SwissProt and GOA datasets. We further demonstrated how our method accurately predicted gene members for a given pathway in Reactome, KEGG and MSigDB only based on the pathway description. Finally, we showed how ProTranslator enabled us to generate the textual description instead of the function label for a set of proteins, providing a new scheme for protein function prediction. We envision ProTranslator will give rise to a protein function "search engine" that returns a list of proteins based on the free text queried by the user.
Proteins are fundamental biological entities mediating key roles in cellular function and disease. This paper introduces a multi-scale graph construction of a protein -- HoloProt -- connecting surface to structure and sequence. The surface captures coarser details of the protein, while sequence as primary component and structure -- comprising secondary and tertiary components -- capture finer details. Our graph encoder then learns a multi-scale representation by allowing each level to integrate the encoding from level(s) below with the graph at that level. We test the learned representation on different tasks, (i.) ligand binding affinity (regression), and (ii.) protein function prediction (classification). On the regression task, contrary to previous methods, our model performs consistently and reliably across different dataset splits, outperforming all baselines on most splits. On the classification task, it achieves a performance close to the top-performing model while using 10x fewer parameters. To improve the memory efficiency of our construction, we segment the multiplex protein surface manifold into molecular superpixels and substitute the surface with these superpixels at little to no performance loss.
Self-supervised learning of graph neural networks (GNNs) aims to learn an accurate representation of the graphs in an unsupervised manner, to obtain transferable representations of them for diverse downstream tasks. Predictive learning and contrastive learning are the two most prevalent approaches for graph self-supervised learning. However, they have their own drawbacks. While the predictive learning methods can learn the contextual relationships between neighboring nodes and edges, they cannot learn global graph-level similarities. Contrastive learning, while it can learn global graph-level similarities, its objective to maximize the similarity between two differently perturbed graphs may result in representations that cannot discriminate two similar graphs with different properties. To tackle such limitations, we propose a framework that aims to learn the exact discrepancy between the original and the perturbed graphs, coined as Discrepancy-based Self-supervised LeArning (D-SLA). Specifically, we create multiple perturbations of the given graph with varying degrees of similarity, and train the model to predict whether each graph is the original graph or the perturbed one. Moreover, we further aim to accurately capture the amount of discrepancy for each perturbed graph using the graph edit distance. We validate our D-SLA on various graph-related downstream tasks, including molecular property prediction, protein function prediction, and link prediction tasks, on which ours largely outperforms relevant baselines.
Self-supervised protein language models have proved their effectiveness in learning the proteins representations. With the increasing computational power, current protein language models pre-trained with millions of diverse sequences can advance the parameter scale from million-level to billion-level and achieve remarkable improvement. However, those prevailing approaches rarely consider incorporating knowledge graphs (KGs), which can provide rich structured knowledge facts for better protein representations. We argue that informative biology knowledge in KGs can enhance protein representation with external knowledge. In this work, we propose OntoProtein, the first general framework that makes use of structure in GO (Gene Ontology) into protein pre-training models. We construct a novel large-scale knowledge graph that consists of GO and its related proteins, and gene annotation texts or protein sequences describe all nodes in the graph. We propose novel contrastive learning with knowledge-aware negative sampling to jointly optimize the knowledge graph and protein embedding during pre-training. Experimental results show that OntoProtein can surpass state-of-the-art methods with pre-trained protein language models in TAPE benchmark and yield better performance compared with baselines in protein-protein interaction and protein function prediction. Code and datasets are available in https://github.com/zjunlp/OntoProtein.
Attention-based deep networks have been successfully applied on textual data in the field of NLP. However, their application on protein sequences poses additional challenges due to the weak semantics of the protein words, unlike the plain text words. These unexplored challenges faced by the standard attention technique include (i) vanishing attention score problem and (ii) high variations in the attention distribution. In this regard, we introduce a novel {\lambda}-scaled attention technique for fast and efficient modeling of the protein sequences that addresses both the above problems. This is used to develop the {\lambda}-scaled attention network and is evaluated for the task of protein function prediction implemented at the protein sub-sequence level. Experiments on the datasets for biological process (BP) and molecular function (MF) showed significant improvements in the F1 score values for the proposed {\lambda}-scaled attention technique over its counterpart approach based on the standard attention technique (+2.01% for BP and +4.67% for MF) and state-of-the-art ProtVecGen-Plus approach (+2.61% for BP and +4.20% for MF). Further, fast convergence (converging in half the number of epochs) and efficient learning (in terms of very low difference between the training and validation losses) were also observed during the training process.
Recent advances in protein function prediction exploit graph-based deep learning approaches to correlate the structural and topological features of proteins with their molecular functions. However, proteins in vivo are not static but dynamic molecules that alter conformation for functional purposes. Here we apply normal mode analysis to native protein conformations and augment protein graphs by connecting edges between dynamically correlated residue pairs. In the multilabel function classification task, our method demonstrates a remarkable performance gain based on this dynamics-informed representation. The proposed graph neural network, ProDAR, increases the interpretability and generalizability of residue-level annotations and robustly reflects structural nuance in proteins. We elucidate the importance of dynamic information in graph representation by comparing class activation maps for hMTH1, nitrophorin, and SARS-CoV-2 receptor binding domain. Our model successfully learns the dynamic fingerprints of proteins and pinpoints the residues of functional impacts, with vast untapped potential for broad biotechnology and pharmaceutical applications.
We propose an effective method to improve Protein Function Prediction (PFP) utilizing hierarchical features of Gene Ontology (GO) terms. Our method consists of a language model for encoding the protein sequence and a Graph Convolutional Network (GCN) for representing GO terms. To reflect the hierarchical structure of GO to GCN, we employ node(GO term)-wise representations containing the whole hierarchical information. Our algorithm shows effectiveness in a large-scale graph by expanding the GO graph compared to previous models. Experimental results show that our method outperformed state-of-the-art PFP approaches.
The capability of accurate prediction of protein functions and properties is essential in the biotechnology industry, e.g. drug development and artificial protein synthesis, etc. The main challenges of protein function prediction are the large label space and the lack of labeled training data. Our method leverages unsupervised sequence embedding and the success of deep convolutional neural network to overcome these challenges. In contrast, most of the existing methods delete the rare protein functions to reduce the label space. Furthermore, some existing methods require additional bio-information (e.g., the 3-dimensional structure of the proteins) which is difficult to be determined in biochemical experiments. Our proposed method significantly outperforms the other methods on the publicly available benchmark using only protein sequences as input. This allows the process of identifying protein functions to be sped up.
Large self-supervised models pretrained on millions of protein sequences have recently gained popularity in generating embeddings of protein sequences for protein function prediction. However, the absence of random baselines makes it difficult to conclude whether pretraining has learned useful information for protein function prediction. Here we show that one-hot encoding and random embeddings, both of which do not require any pretraining, are strong baselines for protein function prediction across 14 diverse sequence-to-function tasks.
Graph classification is a highly impactful task that plays a crucial role in a myriad of real-world applications such as molecular property prediction and protein function prediction.Aiming to handle the new classes with limited labeled graphs, few-shot graph classification has become a bridge of existing graph classification solutions and practical usage.This work explores the potential of metric-based meta-learning for solving few-shot graph classification.We highlight the importance of considering structural characteristics in the solution and propose a novel framework which explicitly considers global structure and local structure of the input graph. An implementation upon GIN, named SMF-GIN, is tested on two datasets, Chembl and TRIANGLES, where extensive experiments validate the effectiveness of the proposed method. The Chembl is constructed to fill in the gap of lacking large-scale benchmark for few-shot graph classification evaluation, which is released together with the implementation of SMF-GIN at: https://github.com/jiangshunyu/SMF-GIN.
With the rapid growth in high-throughput biological sequencing technologies and subsequently the amount of produced omics data, it is essential to develop automated methods to annotate the functionality of unknown genes and proteins. There are developed tools such as AHRD applying known proteins characterization to annotate unknown ones. Some other algorithms such as eggNOG apply orthologous groups of proteins to detect the most probable function. However, while the available tools focus on the detection of the most similar characterization, they are not able to generalize and integrate information from multiple homologs while maintaining accuracy. Here, we devise AFDP, an integrated approach for protein function prediction which benefits from the combination of two available tools, AHRD and eggNOG, to predict the functionality of novel proteins and produce more precise human readable descriptions by applying our stCFExt algorithm. StCFExt creates function descriptions applying available manually curated descriptions in swiss-prot. Using a benchmark dataset we show that the annotations predicted by our approach are more accurate than eggNOG and AHRD annotations.
In recent years, deep learning algorithms have outperformed the state-of-the art methods in several areas thanks to the efficient methods for training and for preventing overfitting, advancement in computer hardware, the availability of vast amount data. The high performance of multi-task deep neural networks in drug discovery has attracted the attention to deep learning algorithms in bioinformatics area. Here, we proposed a hierarchical multi-task deep neural network architecture based on Gene Ontology (GO) terms as a solution to protein function prediction problem and investigated various aspects of the proposed architecture by performing several experiments. First, we showed that there is a positive correlation between performance of the system and the size of training datasets. Second, we investigated whether the level of GO terms on GO hierarchy related to their performance. We showed that there is no relation between the depth of GO terms on GO hierarchy and their performance. In addition, we included all annotations to the training of a set of GO terms to investigate whether including noisy data to the training datasets change the performance of the system. The results showed that including less reliable annotations in training of deep neural networks increased the performance of the low performed GO terms, significantly. We evaluated the performance of the system using hierarchical evaluation method. Mathews correlation coefficient was calculated as 0.75, 0.49 and 0.63 for molecular function, biological process and cellular component categories, respectively. We showed that deep learning algorithms have a great potential in protein function prediction area. We plan to further improve the DEEPred by including other types of annotations from various biological data sources. We plan to construct DEEPred as an open access online tool.
SMISS is a novel web server for protein function prediction. Three different predictors can be selected for different usage. It integrates different sources to improve the protein function prediction accuracy, including the query protein sequence, protein-protein interaction network, gene-gene interaction network, and the rules mined from protein function associations. SMISS automatically switch to ab initio protein function prediction based on the query sequence when there is no homologs in the database. It takes fasta format sequences as input, and several sequences can submit together without influencing the computation speed too much. PHP and Perl are two primary programming language used in the server. The CodeIgniter MVC PHP web framework and Bootstrap front-end framework are used for building the server. It can be used in different platforms in standard web browser, such as Windows, Mac OS X, Linux, and iOS. No plugins are needed for our website. Availability: http://tulip.rnet.missouri.edu/profunc/.
Understanding protein structure-function relationships is a key challenge in computational biology, with applications across the biotechnology and pharmaceutical industries. While it is known that protein structure directly impacts protein function, many functional prediction tasks use only protein sequence. In this work, we isolate protein structure to make functional annotations for proteins in the Protein Data Bank in order to study the expressiveness of different structure-based prediction schemes. We present PersGNN - an end-to-end trainable deep learning model that combines graph representation learning with topological data analysis to capture a complex set of both local and global structural features. While variations of these techniques have been successfully applied to proteins before, we demonstrate that our hybridized approach, PersGNN, outperforms either method on its own as well as a baseline neural network that learns from the same information. PersGNN achieves a 9.3% boost in area under the precision recall curve (AUPR) compared to the best individual model, as well as high F1 scores across different gene ontology categories, indicating the transferability of this approach.
Hierarchical multi-label classification (HMC) is a challenging classification task extending standard multi-label classification problems by imposing a hierarchy constraint on the classes. In this paper, we propose C-HMCNN(h), a novel approach for HMC problems, which, given a network h for the underlying multi-label classification problem, exploits the hierarchy information in order to produce predictions coherent with the constraint and improve performance. We conduct an extensive experimental analysis showing the superior performance of C-HMCNN(h) when compared to state-of-the-art models.
Protein function prediction may be framed as predicting subgraphs (with certain closure properties) of a directed acyclic graph describing the hierarchy of protein functions. Graph neural networks (GNNs), with their built-in inductive bias for relational data, are hence naturally suited for this task. However, in contrast with most GNN applications, the graph is not related to the input, but to the label space. Accordingly, we propose Tail-GNNs, neural networks which naturally compose with the output space of any neural network for multi-task prediction, to provide relationally-reinforced labels. For protein function prediction, we combine a Tail-GNN with a dilated convolutional network which learns representations of the protein sequence, making significant improvement in F_1 score and demonstrating the ability of Tail-GNNs to learn useful representations of labels and exploit them in real-world problem solving.