Pfam Families Research Articles

Investigation of protein family relationships with deep learning.

In this article, we propose a method for finding similarities between Pfam families based on the pre-trained neural network ProtENN2. We use the model ProtENN2 per-residue embeddings to produce new high-dimensional per-family embeddings and develop an approach for calculating inter-family similarity scores based on these embeddings, and evaluate its predictions using structure comparison. We apply our method toPfam annotation by refining clan membership for Pfam families, suggesting both new members of existing clans and potential new clans for future Pfam releases. We investigate some of the failure modes of our approach, which suggests directions for future improvements. Our method is relatively simple with few parameters and could be applied to other protein family classification models. Overall, our work suggests potential benefits of employing deep learning for improving our understanding of protein family relationships and functions of previously uncharacterized families. github.com/iponamareva/ProtCNNSim, 10.5281/zenodo.10091909.

Bridging the Gap between Sequence and Structure Classifications of Proteins with AlphaFold Models

Classification of protein domains based on homology and structural similarity serves as a fundamental tool to gain biological insights into protein function. Recent advancements in protein structure prediction, exemplified by AlphaFold, have revolutionized the availability of protein structural data. We focus on classifying about 9000 Pfam families into ECOD (Evolutionary Classification of Domains) by using predicted AlphaFold models and the DPAM (Domain Parser for AlphaFold Models) tool. Our results offer insights into their homologous relationships and domain boundaries. More than half of these Pfam families contain DPAM domains that can be confidently assigned to the ECOD hierarchy. Most assigned domains belong to highly populated folds such as Immunoglobulin-like (IgL), Armadillo (ARM), helix-turn-helix (HTH), and Src homology 3 (SH3). A large fraction of DPAM domains, however, cannot be confidently assigned to ECOD homologous groups. These unassigned domains exhibit statistically different characteristics, including shorter average length, fewer secondary structure elements, and more abundant transmembrane segments. They could potentially define novel families remotely related to domains with known structures or novel superfamilies and folds. Manual scrutiny of a subset of these domains revealed an abundance of internal duplications and recurring structural motifs. Exploring sequence and structural features such as disulfide bond patterns, metal-binding sites, and enzyme active sites helped uncover novel structural folds as well as remote evolutionary relationships. By bridging the gap between sequence-based Pfam and structure-based ECOD domain classifications, our study contributes to a more comprehensive understanding of the protein universe by providing structural and functional insights into previously uncharacterized proteins.

Deciphering the Diversity in Bacterial Transporters That Salvage Queuosine Precursors.

Queuosine (Q) is a modification of the wobble base of tRNA harboring GUN anticodons with roles in decoding accuracy and efficiency. Its synthesis is complex with multiple enzymatic steps, and several pathway intermediates can be salvaged. The only two transporter families known to salvage Q precursors are QPTR/COG1738 and QrtT/QueT. Analyses of the distribution of known Q synthesis and salvage genes in human gut and oral microbiota genomes have suggested that more transporter families remain to be found and that Q precursor exchanges must occur within the structured microenvironments of the mammalian host. Using physical clustering and fusion-based association with Q salvage genes, candidate genes for missing transporters were identified and five were tested experimentally by complementation assays in Escherichia coli. Three genes encoding transporters from three different Pfam families, a ureide permease (PF07168) from Acidobacteriota bacterium, a hemolysin III family protein (PF03006) from Bifidobacterium breve, and a Major Facilitator Superfamily protein (PF07690) from Bartonella henselae, were found to allow the transport of both preQ0 and preQ1 in this heterologous system. This work suggests that many transporter families can evolve to transport Q precursors, reinforcing the concept of transporter plasticity.

Evaluating large language models for annotating proteins.

In UniProtKB, up to date, there are more than 251 million proteins deposited. However, only 0.25% have been annotated with one of the more than 15000 possible Pfam family domains. The current annotation protocol integrates knowledge from manually curated family domains, obtained using sequence alignments and hidden Markov models. This approach has been successful for automatically growing the Pfam annotations, however at a low rate in comparison to protein discovery. Just a few years ago, deep learning models were proposed for automatic Pfam annotation. However, these models demand a considerable amount of training data, which can be a challenge with poorly populated families. To address this issue, we propose and evaluate here a novel protocol based on transfer learningṪhis requires the use of protein large language models (LLMs), trained with self-supervision on big unnanotated datasets in order to obtain sequence embeddings. Then, the embeddings can be used with supervised learning on a small and annotated dataset for a specialized task. In this protocol we have evaluated several cutting-edge protein LLMs together with machine learning architectures to improve the actual prediction of protein domain annotations. Results are significatively better than state-of-the-art for protein families classification, reducing the prediction error by an impressive 60% compared to standard methods. We explain how LLMs embeddings can be used for protein annotation in a concrete and easy way, and provide the pipeline in a github repo. Full source code and data are available at https://github.com/sinc-lab/llm4pfam.

In vitro characterization of alternative l-threonate and d-erythronate catabolic pathways

l-threonate is the metabolite of vitamin C, while d-erythronate is the metabolite of N-acetyl-d-glucosamine, the nutritional supplement for joint health. They are widely distributed in the environment and human biofluids. Nevertheless, the catabolisms of l-threonate and d-erythronate are sparsely reported. Here we explored the functional diversity of an acid sugar kinase family (Pfam families PF07005-PF17042), and discovered a novel 2-oxo-tetronate kinase. The conserved genome neighborhood of the 2-oxo-tetronate kinase encodes members of class-II fructose-bisphosphate aldolase family (F_bP_aldolase, PF01116) and a dehydrogenase family (PF03446-PF14833). Instructed by this analysis, we experimentally verified that these enzymes are capable of degrading l-threonate into dihydroxyacetone phosphate (DHAP) in Arthrobacter sp. ZBG10, Clostridium scindens ATCC 35704, and Pseudonocardia dioxanivorans ATCC 55486. Meanwhile, a convergent catabolic pathway for d-erythronate was characterized in P. dioxanivorans ATCC 55486. Moreover, the phylogenetic distribution analysis indicates that the biological range of the identified l-threonate and d-erythronate catabolic pathways appears to extend mostly to members of the Actinomycetota, Cyanobacteriota, Bacillota, Pseudomonadota, and Bacteroidota phyla.

Drying without dying: A genome database for desiccation-tolerant plants and evolution of desiccation tolerance.

Desiccation is typically fatal, but a small number of land plants have evolved vegetative desiccation tolerance (VDT), allowing them to dry without dying through a process called anhydrobiosis. Advances in sequencing technologies have enabled the investigation of genomes for desiccation-tolerant plants over the past decade. However, a dedicated and integrated database for these valuable genomic resources has been lacking. Our prolonged interest in VDT plant genomes motivated us to create the "Drying without Dying" database, which contains a total of 16 VDT-related plant genomes (including 10 mosses) and incorporates 10 genomes that are closely related to VDT plants. The database features bioinformatic tools, such as blast and homologous cluster search, sequence retrieval, Gene Ontology term and metabolic pathway enrichment statistics, expression profiling, co-expression network extraction, and JBrowser exploration for each genome. To demonstrate its utility, we conducted tailored PFAM family statistical analyses, and we discovered that the drought-responsive ABA transporter AWPM-19 family is significantly tandemly duplicated in all bryophytes but rarely so in tracheophytes. Transcriptomic investigations also revealed that response patterns following desiccation diverged between bryophytes and angiosperms. Combined, the analyses provided genomic and transcriptomic evidence supporting a possible divergence and lineage-specific evolution of VDT in plants. The database can be accessed at http://desiccation.novogene.com. We expect this initial release of the "Drying without Dying" plant genome database will facilitate future discovery of VDT genetic resources.

Sensing preferences for prokaryotic solute binding protein families.

Solute binding proteins (SBPs) are of central physiological relevance for prokaryotes. These proteins present substrates to transporters, but they also stimulate different signal transduction receptors. SBPs form a superfamily of at least 33 protein Pfam families. To assess possible links between SBP sequence and the ligand recognized, we have inspected manually all SBP three-dimensional structures deposited in the protein data bank and retrieved 748 prokaryotic structures that have been solved in complex with bound ligand. These structures were classified into 26 SBP Pfam families. The analysis of the ligands recognized revealed that most families possess a preference for a compound class. There were three families each that bind preferentially saccharides and amino acids. In addition, we identified families that bind preferentially purines, quaternary amines, iron and iron-chelating compounds, oxoanions, bivalent metal ions or phosphates. Phylogenetic analyses suggest convergent evolutionary events that lead to families that bind the same ligand. The functional link between chemotaxis and compound uptake is reflected in similarities in the ligands recognized by SBPs and chemoreceptors. Associating Pfam families with ligand profiles will be of help to design experimental strategies aimed at the identification of ligands for uncharacterized SBPs.

Implementation of machine learning models to determine the appropriate model for protein function prediction

Predicting the function of proteins is a crucial part of genome annotation, which can help in solving a wide range of biological problems. Many methods are available to predict the functions of proteins. However, except for sequence, most features are difficult to obtain or are not available for many proteins, which limits their scope. In addition, the performance of sequence-based feature prediction methods is often lower than that of methods that involve multiple features, and protein feature prediction can be time-consuming. Recent advances in this field are associated with the development of machine learning, which shows great progress in solving the problem of predicting protein functions. Today, however, most protein sequences have the status of «uncharacterized» or «putative». The need to assess the accuracy of identification of protein functions is an urgent task for machine learning approaches used to predict protein functions. In this study, the performance of two popular function prediction algorithms (ProtCNN and BiLSTM) was assessed from two perspectives and the procedures for building these models were described. As a result of the study of Pfam families, ProtCNN achieves an accuracy rate of 0.988 % and bidirectional LSTM has an accuracy rate of 0.9506 %. The use of the Pfam dataset allowed increasing the classification accuracy due to the large training dataset. The quality of the prediction increases with a large amount of training data. The study demonstrated that machine learning algorithms can be used as an effective tool for building protein function prediction models, in particular, the CNN network can be adapted as an accurate tool for annotating protein functions in the presence of large datasets.

DPCfam: Unsupervised protein family classification by Density Peak Clustering of large sequence datasets.

Proteins that are known only at a sequence level outnumber those with an experimental characterization by orders of magnitude. Classifying protein regions (domains) into homologous families can generate testable functional hypotheses for yet unannotated sequences. Existing domain family resources typically use at least some degree of manual curation: they grow slowly over time and leave a large fraction of the protein sequence space unclassified. We here describe automatic clustering by Density Peak Clustering of UniRef50 v. 2017_07, a protein sequence database including approximately 23M sequences. We performed a radical re-implementation of a pipeline we previously developed in order to allow handling millions of sequences and data volumes of the order of 3 TeraBytes. The modified pipeline, which we call DPCfam, finds ∼ 45,000 protein clusters in UniRef50. Our automatic classification is in close correspondence to the ones of the Pfam and ECOD resources: in particular, about 81% of medium-large Pfam families and 72% of ECOD families can be mapped to clusters generated by DPCfam. In addition, our protocol finds more than 14,000 clusters constituted of protein regions with no Pfam annotation, which are therefore candidates for representing novel protein families. These results are made available to the scientific community through a dedicated repository.

Transcription Factors in Aureobasidium spp.: Classification, Regulation and a Newly Built Database.

Transcription factors (TFs) can regulate the synthesis of secondary metabolites through different metabolic pathways in Aureobasidium spp. In this study, a set of 16 superfamilies, 45 PFAM families of TFs with the DNA-binding domains, seven zinc finger families and eight categories of the C2H2 TFs have been identified in Aureobasidium spp. Among all the identified TFs, four superfamilies and six PFAM families are the fungal-specific types in this lineage. The Zn2Cys6 and fungal-specific domain regulators are found to be overwhelmingly predominated, while the C2H2 zinc finger class comprises a smaller regulator class. Since there are currently no databases that allow for easy exploration of the TFs in Aureobasidium spp., based on over 50 references and 2405 homologous TFs, the first TFs pipeline—the Aureobasidium Transcription Factor Database (ATFDB)—has been developed to accelerate the identification of metabolic regulation in various Aureobasidium species. It would be useful to investigate the mechanisms behind the wide adaptability and metabolite diversity of Aureobasidium spp.

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

Deep learning-based prediction of protein structure usually begins by constructing a multiple sequence alignment (MSA) containing homologs of the target protein. The most successful approaches combine large feature sets derived from MSAs, and considerable computational effort is spent deriving these input features. We present a method that greatly reduces the amount of preprocessing required for a target MSA, while producing main chain coordinates as a direct output of a deep neural network. The network makes use of just three recurrent networks and a stack of residual convolutional layers, making the predictor very fast to run, and easy to install and use. Our approach constructs a directly learned representation of the sequences in an MSA, starting from a one-hot encoding of the sequences. When supplemented with an approximate precision matrix, the learned representation can be used to produce structural models of comparable or greater accuracy as compared to our original DMPfold method, while requiring less than a second to produce a typical model. This level of accuracy and speed allows very large-scale three-dimensional modeling of proteins on minimal hardware, and we demonstrate this by producing models for over 1.3 million uncharacterized regions of proteins extracted from the BFD sequence clusters. After constructing an initial set of approximate models, we select a confident subset of over 30,000 models for further refinement and analysis, revealing putative novel protein folds. We also provide updated models for over 5,000 Pfam families studied in the original DMPfold paper.

PreBINDS: An Interactive Web Tool to Create Appropriate Datasets for Predicting Compound–Protein Interactions

Given the abundant computational resources and the huge amount of data of compound–protein interactions (CPIs), constructing appropriate datasets for learning and evaluating prediction models for CPIs is not always easy. For this study, we have developed a web server to facilitate the development and evaluation of prediction models by providing an appropriate dataset according to the task. Our web server provides an environment and dataset that aid model developers and evaluators in obtaining a suitable dataset for both proteins and compounds, in addition to attributes necessary for deep learning. With the web server interface, users can customize the CPI dataset derived from ChEMBL by setting positive and negative thresholds to be adjusted according to the user’s definitions. We have also implemented a function for graphic display of the distribution of activity values in the dataset as a histogram to set appropriate thresholds for positive and negative examples. These functions enable effective development and evaluation of models. Furthermore, users can prepare their task-specific datasets by selecting a set of target proteins based on various criteria such as Pfam families, ChEMBL’s classification, and sequence similarities. The accuracy and efficiency of in silico screening and drug design using machine learning including deep learning can therefore be improved by facilitating access to an appropriate dataset prepared using our web server (https://binds.lifematics.work/).

SCOPe: improvements to the structural classification of proteins - extended database to facilitate variant interpretation and machine learning.

The Structural Classification of Proteins—extended (SCOPe, https://scop.berkeley.edu) knowledgebase aims to provide an accurate, detailed, and comprehensive description of the structural and evolutionary relationships amongst the majority of proteins of known structure, along with resources for analyzing the protein structures and their sequences. Structures from the PDB are divided into domains and classified using a combination of manual curation and highly precise automated methods. In the current release of SCOPe, 2.08, we have developed search and display tools for analysis of genetic variants we mapped to structures classified in SCOPe. In order to improve the utility of SCOPe to automated methods such as deep learning classifiers that rely on multiple alignment of sequences of homologous proteins, we have introduced new machine-parseable annotations that indicate aberrant structures as well as domains that are distinguished by a smaller repeat unit. We also classified structures from 74 of the largest Pfam families not previously classified in SCOPe, and we improved our algorithm to remove N- and C-terminal cloning, expression and purification sequences from SCOPe domains. SCOPe 2.08-stable classifies 106 976 PDB entries (about 60% of PDB entries).

The trRosetta server for fast and accurate protein structure prediction.

The trRosetta (transform-restrained Rosetta) server is a web-based platform for fast and accurate protein structure prediction, powered by deep learning and Rosetta. With the input of a protein's amino acid sequence, a deep neural network is first used to predict the inter-residue geometries, including distance and orientations. The predicted geometries are then transformed as restraints to guide the structure prediction on the basis of direct energy minimization, which is implemented under the framework of Rosetta. The trRosetta server distinguishes itself from other similar structure prediction servers in terms of rapid and accurate de novo structure prediction. As an illustration, trRosetta was applied to two Pfam families with unknown structures, for which the predicted de novo models were estimated to have high accuracy. Nevertheless, to take advantage of homology modeling, homologous templates are used as additional inputs to the network automatically. In general, it takes ~1 h to predict the final structure for a typical protein with ~300 amino acids, using a maximum of 10 CPU cores in parallel in our cluster system. To enable large-scale structure modeling, a downloadable package of trRosetta with open-source codes is available as well. A detailed guidance for using the package is also available in this protocol. The server and the package are available at https://yanglab.nankai.edu.cn/trRosetta/ and https://yanglab.nankai.edu.cn/trRosetta/download/ , respectively.

A catalogue of signal molecules that interact with sensor kinases, chemoreceptors and transcriptional regulators.

Bacteria have evolved many different signal transduction systems that sense signals and generate a variety of responses. Generally, most abundant are transcriptional regulators, sensor histidine kinases and chemoreceptors. Typically, these systems recognize their signal molecules with dedicated ligand-binding domains (LBDs), which, in turn, generate a molecular stimulus that modulates the activity of the output module. There are an enormous number of different LBDs that recognize a similarly diverse set of signals. To give a global perspective of the signals that interact with transcriptional regulators, sensor kinases and chemoreceptors, we manually retrieved information on the protein-ligand interaction from about 1,200 publications and 3D structures. The resulting 811 proteins were classified according to the Pfam family into 127 groups. These data permit a delineation of the signal profiles of individual LBD families as well as distinguishing between families that recognize signals in a promiscuous manner and those that possess a well-defined ligand range. A major bottleneck in the field is the fact that the signal input of many signaling systems is unknown. The signal repertoire reported here will help the scientific community design experimental strategies to identify the signaling molecules for uncharacterised sensor proteins.

Folding non-homologous proteins by coupling deep-learning contact maps with I-TASSER assembly simulations

SummaryStructure prediction for proteins lacking homologous templates in the Protein Data Bank (PDB) remains a significant unsolved problem. We developed a protocol, C-I-TASSER, to integrate interresidue contact maps from deep neural-network learning with the cutting-edge I-TASSER fragment assembly simulations. Large-scale benchmark tests showed that C-I-TASSER can fold more than twice the number of non-homologous proteins than the I-TASSER, which does not use contacts. When applied to a folding experiment on 8,266 unsolved Pfam families, C-I-TASSER successfully folded 4,162 domain families, including 504 folds that are not found in the PDB. Furthermore, it created correct folds for 85% of proteins in the SARS-CoV-2 genome, despite the quick mutation rate of the virus and sparse sequence profiles. The results demonstrated the critical importance of coupling whole-genome and metagenome-based evolutionary information with optimal structure assembly simulations for solving the problem of non-homologous protein structure prediction.

Evolution and insights into the structure and function of the DedA superfamily containing TMEM41B and VMP1.

TMEM41B and VMP1 are endoplasmic reticulum (ER)-localizing multi-spanning membrane proteins required for ER-related cellular processes such as autophagosome formation, lipid droplet homeostasis and lipoprotein secretion in eukaryotes. Both proteins have a VTT domain, which is similar to the DedA domain found in bacterial DedA family proteins. However, the molecular function and structure of the DedA and VTT domains (collectively referred to as DedA domains) and the evolutionary relationships among the DedA domain-containing proteins are largely unknown. Here, we conduct a remote homology search and identify a new clade consisting mainly of bacterial proteins of unknown function that are members of the Pfam family PF06695. Phylogenetic analysis reveals that the TMEM41, VMP1, DedA and PF06695 families form a superfamily with a common origin, which we term the DedA superfamily. Coevolution-based structural prediction suggests that the DedA domain contains two reentrant loops facing each other in the membrane. This topology is biochemically verified by the substituted cysteine accessibility method. The predicted structure is topologically similar to that of the substrate-binding region of Na+-coupled glutamate transporter solute carrier 1 (SLC1) proteins. A potential ion-coupled transport function of the DedA superfamily proteins is discussed. This article has an associated First Person interview with the joint first authors of the paper.

Accurate contact-based modelling of repeat proteins predicts the structure of new repeats protein families.

Repeat proteins are abundant in eukaryotic proteomes. They are involved in many eukaryotic specific functions, including signalling. For many of these proteins, the structure is not known, as they are difficult to crystallise. Today, using direct coupling analysis and deep learning it is often possible to predict a protein's structure. However, the unique sequence features present in repeat proteins have been a challenge to use direct coupling analysis for predicting contacts. Here, we show that deep learning-based methods (trRosetta, DeepMetaPsicov (DMP) and PconsC4) overcomes this problem and can predict intra- and inter-unit contacts in repeat proteins. In a benchmark dataset of 815 repeat proteins, about 90% can be correctly modelled. Further, among 48 PFAM families lacking a protein structure, we produce models of forty-one families with estimated high accuracy.

Density Peak clustering of protein sequences associated to a Pfam clan reveals clear similarities and interesting differences with respect to manual family annotation

BackgroundThe identification of protein families is of outstanding practical importance for in silico protein annotation and is at the basis of several bioinformatic resources. Pfam is possibly the most well known protein family database, built in many years of work by domain experts with extensive use of manual curation. This approach is generally very accurate, but it is quite time consuming and it may suffer from a bias generated from the hand-curation itself, which is often guided by the available experimental evidence.ResultsWe introduce a procedure that aims to identify automatically putative protein families. The procedure is based on Density Peak Clustering and uses as input only local pairwise alignments between protein sequences. In the experiment we present here, we ran the algorithm on about 4000 full-length proteins with at least one domain classified by Pfam as belonging to the Pseudouridine synthase and Archaeosine transglycosylase (PUA) clan. We obtained 71 automatically-generated sequence clusters with at least 100 members. While our clusters were largely consistent with the Pfam classification, showing good overlap with either single or multi-domain Pfam family architectures, we also observed some inconsistencies. The latter were inspected using structural and sequence based evidence, which suggested that the automatic classification captured evolutionary signals reflecting non-trivial features of protein family architectures. Based on this analysis we identified a putative novel pre-PUA domain as well as alternative boundaries for a few PUA or PUA-associated families. As a first indication that our approach was unlikely to be clan-specific, we performed the same analysis on the P53 clan, obtaining comparable results.ConclusionsThe clustering procedure described in this work takes advantage of the information contained in a large set of pairwise alignments and successfully identifies a set of putative families and family architectures in an unsupervised manner. Comparison with the Pfam classification highlights significant overlap and points to interesting differences, suggesting that our new algorithm could have potential in applications related to automatic protein classification. Testing this hypothesis, however, will require further experiments on large and diverse sequence datasets.

Pfam: The protein families database in 2021.

The Pfam database is a widely used resource for classifying protein sequences into families and domains. Since Pfam was last described in this journal, over 350 new families have been added in Pfam 33.1 and numerous improvements have been made to existing entries. To facilitate research on COVID-19, we have revised the Pfam entries that cover the SARS-CoV-2 proteome, and built new entries for regions that were not covered by Pfam. We have reintroduced Pfam-B which provides an automatically generated supplement to Pfam and contains 136 730 novel clusters of sequences that are not yet matched by a Pfam family. The new Pfam-B is based on a clustering by the MMseqs2 software. We have compared all of the regions in the RepeatsDB to those in Pfam and have started to use the results to build and refine Pfam repeat families. Pfam is freely available for browsing and download at http://pfam.xfam.org/.