scholarly journals Deep neural language modeling enables functional protein generation across families

2021 ◽  
Author(s):  
Ali Madani ◽  
Ben Krause ◽  
Eric R Greene ◽  
Subu Subramanian ◽  
Benjamin P Mohr ◽  
...  
Keyword(s):  
Amino Acid ◽  
De Novo ◽  
Language Model ◽  
Protein Sequences ◽  
Chorismate Mutase ◽  
Language Models ◽  
Protein Families ◽  
Evolutionary Trajectory ◽  
Artificial Proteins ◽  
Artificial Protein

Bypassing nature's evolutionary trajectory, de novo protein generation - defined as creating artificial protein sequences from scratch - could enable breakthrough solutions for biomedical and environmental challenges. Viewing amino acid sequences as a language, we demonstrate that a deep learning-based language model can generate functional artificial protein sequences across families, akin to generating grammatically and semantically correct natural language sentences on diverse topics. Our protein language model is trained by simply learning to predict the next amino acid for over 280 million protein sequences from thousands of protein families, without biophysical or coevolutionary modeling. We experimentally evaluate model-generated artificial proteins on five distinct antibacterial lysozyme families. Artificial proteins show similar activities and catalytic efficiencies as representative natural lysozymes, including hen egg white lysozyme, while reaching as low as 44% identity to any known naturally-evolved protein. The X-ray crystal structure of an enzymatically active artificial protein recapitulates the conserved fold and positioning of active site residues found in natural proteins. We demonstrate our language model's ability to be adapted to different protein families by accurately predicting the functionality of artificial chorismate mutase and malate dehydrogenase proteins. These results indicate that neural language models successfully perform de novo protein generation across protein families and may prove to be a tool to shortcut evolution.

scholarly journals Unevolved De Novo Proteins Have Innate Tendencies to Bind Transition Metals

Life ◽  
2019 ◽  
Vol 9 (1) ◽  
pp. 8 ◽  
Author(s):  
Michael S. Wang ◽  
Kenric J. Hoegler ◽  
Michael H. Hecht
Keyword(s):  
Amino Acid ◽  
Transition Metals ◽  
Metal Binding ◽  
Combinatorial Library ◽  
De Novo ◽  
Protein Sequences ◽  
Amino Acid Sequences ◽  
Ancestral Sequences ◽  
Wide Range ◽  
Catalytic Functions

Life as we know it would not exist without the ability of protein sequences to bind metal ions. Transition metals, in particular, play essential roles in a wide range of structural and catalytic functions. The ubiquitous occurrence of metalloproteins in all organisms leads one to ask whether metal binding is an evolved trait that occurred only rarely in ancestral sequences, or alternatively, whether it is an innate property of amino acid sequences, occurring frequently in unevolved sequence space. To address this question, we studied 52 proteins from a combinatorial library of novel sequences designed to fold into 4-helix bundles. Although these sequences were neither designed nor evolved to bind metals, the majority of them have innate tendencies to bind the transition metals copper, cobalt, and zinc with high nanomolar to low-micromolar affinity.

scholarly journals In silico analysis of virulence associated genes in genomes of Escherichia coli strains causing colibacillosis in poultry

2017 ◽  
Vol 61 (4) ◽  
pp. 421-426 ◽  
Author(s):  
Joanna Kołsut ◽  
Paulina Borówka ◽  
Błażej Marciniak ◽  
Ewelina Wójcik ◽  
Arkadiusz Wojtasik ◽  
...  
Keyword(s):  
Escherichia Coli ◽  
Amino Acid ◽  
Virulence Factors ◽  
De Novo ◽  
Protein Sequences ◽  
In Silico Analysis ◽  
Amino Acid Sequences ◽  
Common Disease ◽  
Bacterial Genomes ◽  
E Coli

AbstractIntroduction: Colibacillosis – the most common disease of poultry, is caused mainly by avian pathogenic Escherichia coli (APEC). However, thus far, no pattern to the molecular basis of the pathogenicity of these bacteria has been established beyond dispute. In this study, genomes of APEC were investigated to ascribe importance and explore the distribution of 16 genes recognised as their virulence factors.Material and Methods: A total of 14 pathogenic for poultry E. coli strains were isolated, and their DNA was sequenced, assembled de novo, and annotated. Amino acid sequences from these bacteria and an additional 16 freely available APEC amino acid sequences were analysed with the DIFFIND tool to define their virulence factors.Results: The DIFFIND tool enabled quick, reliable, and convenient assessment of the differences between compared amino acid sequences from bacterial genomes. The presence of 16 protein sequences indicated as pathogenicity factors in poultry resulted in the generation of a heatmap which categorises genomes in terms of the existence and similarity of the analysed protein sequences.Conclusion: The proposed method of detection of virulence factors using the capabilities of the DIFFIND tool may be useful in the analysis of similarities of E. coli and other sequences deriving from bacteria. Phylogenetic analysis resulted in reliable segregation of 30 APEC strains into five main clusters containing various virulence associated genes (VAGs).

scholarly journals Phylogenetic correlations have limited effect on coevolution-based contact prediction in proteins

2020 ◽  
Author(s):  
Edwin Rodriguez Horta ◽  
Martin Weigt
Keyword(s):  
Protein Structure ◽  
Amino Acid ◽  
Protein Structure Prediction ◽  
Structure Prediction ◽  
Protein Sequences ◽  
Protein Families ◽  
Coupling Analysis ◽  
Contact Prediction ◽  
Phylogenetic Relations ◽  
Direct Coupling Analysis

AbstractCoevolution-based contact prediction, either directly by coevolutionary couplings resulting from global statistical sequence models or using structural supervision and deep learning, has found widespread application in protein-structure prediction from sequence. However, one of the basic assumptions in global statistical modeling is that sequences form an at least approximately independent sample of an unknown probability distribution, which is to be learned from data. In the case of protein families, this assumption is obviously violated by phylogenetic relations between protein sequences. It has turned out to be notoriously difficult to take phylogenetic correlations into account in coevolutionary model learning. Here, we propose a complementary approach: we develop two strategies to randomize or resample sequence data, such that conservation patterns and phylogenetic relations are preserved, while intrinsic (i.e. structure- or function-based) coevolutionary couplings are removed. An analysis of these data shows that the strongest coevolutionary couplings, i.e. those used by Direct Coupling Analysis to predict contacts, are only weakly influenced by phylogeny. However, phylogeny-induced spurious couplings are of similar size to the bulk of coevolutionary couplings, and dissecting functional from phylogeny-induced couplings might lead to more accurate contact predictions in the range of intermediate-size couplings.The code is available at https://github.com/ed-rodh/Null_models_I_and_II.Author summaryMany homologous protein families contain thousands of highly diverged amino-acid sequences, which fold in close-to-identical three-dimensional structures and fulfill almost identical biological tasks. Global coevolutionary models, like those inferred by the Direct Coupling Analysis (DCA), assume that families can be considered as samples of some unknown statistical model, and that the parameters of these models represent evolutionary constraints acting on protein sequences. To learn these models from data, DCA and related approaches have to also assume that the distinct sequences in a protein family are close to independent, while in reality they are characterized by involved hierarchical phylogenetic relationships. Here we propose Null models for sequence alignments, which maintain patterns of amino-acid conservation and phylogeny contained in the data, but destroy any coevolutionary couplings, frequently used in protein structure prediction. We find that phylogeny actually induces spurious non-zero couplings. These are, however, significantly smaller that the largest couplings derived from natural sequences, and therefore have only little influence on the first predicted contacts. However, in the range of intermediate couplings, they may lead to statistically significant effects. Dissecting phylogenetic from functional couplings might therefore extend the range of accurately predicted structural contacts down to smaller coupling strengths than those currently used.

scholarly journals The power of universal contextualised protein embeddings in cross-species protein function prediction

2021 ◽  
Author(s):  
Irene van den Bent ◽  
Stavros Makrodimitris ◽  
Marcel Reinders
Keyword(s):  
Amino Acid ◽  
Protein Function ◽  
Prediction Models ◽  
Protein Sequences ◽  
Function Prediction ◽  
Training Data ◽  
Language Models ◽  
Molecular Function ◽  
Test Species ◽  
Protein Functions

AbstractComputationally annotating proteins with a molecular function is a difficult problem that is made even harder due to the limited amount of available labelled protein training data. A recently published supervised molecular function predicting model partly circumvents this limitation by making its predictions based on the universal (i.e. task-agnostic) contextualised protein embeddings from the deep pre-trained unsupervised protein language model SeqVec. SeqVec embeddings incorporate contextual information of amino acids, thereby modelling the underlying principles of protein sequences insensitive to the context of species.We applied the existing SeqVec-based molecular function prediction model in a transfer learning task by training the model on annotated protein sequences of one training species and making predictions on the proteins of several test species with varying evolutionary distance. We show that this approach successfully generalises knowledge about protein function from one eukaryotic species to various other species, proving itself an effective method for molecular function prediction in inadequately annotated species from understudied taxonomic kingdoms. Furthermore, we submitted the performance of our SeqVec-based prediction models to detailed characterisation, first to advance the understanding of protein language models and second to determine areas of improvement.Author summaryProteins are diverse molecules that regulate all processes in biology. The field of synthetic biology aims to understand these protein functions to solve problems in medicine, manufacturing, and agriculture. Unfortunately, for many proteins only their amino acid sequence is known whereas their function remains unknown. Only a few species have been well-studied such as mouse, human and yeast. Hence, we need to increase knowledge on protein functions. Doing so is, however, complicated as determining protein functions experimentally is time-consuming, expensive, and technically limited. Computationally predicting protein functions offers a faster and more scalable approach but is hampered as it requires much data to design accurate function prediction algorithms. Here, we show that it is possible to computationally generalize knowledge on protein function from one well-studied training species to another test species. Additionally, we show that the quality of these protein function predictions depends on how structurally similar the proteins are between the species. Advantageously, the predictors require only the annotations of proteins from the training species and mere amino acid sequences of test species which may particularly benefit the function prediction of species from understudied taxonomic kingdoms such as the Plantae, Protozoa and Chromista.

Protein Organization with Manifold Exploration and Spectral Clustering

2021 ◽  
Author(s):  
Geoffroy Dubourg-Felonneau ◽  
Shahab Shams ◽  
Eyal Akiva ◽  
Lawrence Lee
Keyword(s):  
Spectral Clustering ◽  
Protein Sequences ◽  
Language Models ◽  
Functional Categories ◽  
Protein Families ◽  
Enzyme Protein ◽  
Amount Of Information ◽  
Vast Amount ◽  
Unsupervised Approach

We present a method to provide a biologically meaningful representation of the space of protein sequences. While billions of protein sequences are available, organizing this vast amount of information into functional categories is daunting, time-consuming and incomplete. We present our unsupervised approach that combines Transformer protein language models, UMAP graphs, and spectral clustering to create meaningful clusters in the protein spaces. To demonstrate the meaningfulness of the clusters, we show that they preserve most of the signal present in a dataset of manually curated enzyme protein families.

scholarly journals Generating novel protein sequences using Gibbs sampling of masked language models

2021 ◽  
Author(s):  
Sean R. Johnson ◽  
Sarah Monaco ◽  
Kenneth Massie ◽  
Zaid Syed
Keyword(s):  
Gibbs Sampling ◽  
Protein Sequences ◽  
Generative Models ◽  
Chorismate Mutase ◽  
Language Models ◽  
Chemical Production ◽  
Specific Data ◽  
Natural Language Text ◽  
Novel Protein

AbstractRecently developed language models (LMs) based on deep neural networks have demonstrated the ability to generate fluent natural language text. LMs pre-trained on protein sequences have shown state of the art performance on a variety of downstream tasks. Protein LMs have also been used to generate novel protein sequences. In the present work we use Gibbs sampling of BERT-style LMs, pre-trained on protein sequences using the masked language modeling task, to generate novel protein sequences. We evaluate the quality of the generated sequences by comparing them to natural sequences from the same family. In particular, we focus on proteins from the chorismate mutase type II family, which has been used in previous work as an example target for protein generative models. We find that the Gibbs sampling process on BERT-style models pretrained on millions to billions of protein sequences is able to generate novel sequences that retain key features of related natural sequences. Further, we find that smaller models fine-tuned or trained from scratch on family-specific data are able to equal or surpass the generation quality of large pre-trained models by some metrics. The ability to generate novel natural-like protein sequences could contribute to the development of improved protein therapeutics and protein-catalysts for industrial chemical production.

Computational Analysis of Therapeutic Enzyme Uricase from Different Source Organisms

2020 ◽  
Vol 17 (1) ◽  
pp. 59-77
Author(s):  
Anand Kumar Nelapati ◽  
JagadeeshBabu PonnanEttiyappan
Keyword(s):  
Uric Acid ◽  
Amino Acid ◽  
Sequence Alignment ◽  
Multiple Sequence Alignment ◽  
Protein Sequences ◽  
Amino Acid Sequences ◽  
Amino Acid Residues ◽  
Multiple Sequence ◽  
Physiochemical Properties ◽  
Pharmaceutical Industries

Background:Hyperuricemia and gout are the conditions, which is a response of accumulation of uric acid in the blood and urine. Uric acid is the product of purine metabolic pathway in humans. Uricase is a therapeutic enzyme that can enzymatically reduces the concentration of uric acid in serum and urine into more a soluble allantoin. Uricases are widely available in several sources like bacteria, fungi, yeast, plants and animals.Objective:The present study is aimed at elucidating the structure and physiochemical properties of uricase by insilico analysis.Methods:A total number of sixty amino acid sequences of uricase belongs to different sources were obtained from NCBI and different analysis like Multiple Sequence Alignment (MSA), homology search, phylogenetic relation, motif search, domain architecture and physiochemical properties including pI, EC, Ai, Ii, and were performed.Results:Multiple sequence alignment of all the selected protein sequences has exhibited distinct difference between bacterial, fungal, plant and animal sources based on the position-specific existence of conserved amino acid residues. The maximum homology of all the selected protein sequences is between 51-388. In singular category, homology is between 16-337 for bacterial uricase, 14-339 for fungal uricase, 12-317 for plants uricase, and 37-361 for animals uricase. The phylogenetic tree constructed based on the amino acid sequences disclosed clusters indicating that uricase is from different source. The physiochemical features revealed that the uricase amino acid residues are in between 300- 338 with a molecular weight as 33-39kDa and theoretical pI ranging from 4.95-8.88. The amino acid composition results showed that valine amino acid has a high average frequency of 8.79 percentage compared to different amino acids in all analyzed species.Conclusion:In the area of bioinformatics field, this work might be informative and a stepping-stone to other researchers to get an idea about the physicochemical features, evolutionary history and structural motifs of uricase that can be widely used in biotechnological and pharmaceutical industries. Therefore, the proposed in silico analysis can be considered for protein engineering work, as well as for gout therapy.

iAFP-gap-SMOTE: An Efficient Feature Extraction Scheme Gapped Dipeptide Composition is Coupled with an Oversampling Technique for Identification of Antifreeze Proteins

2019 ◽  
Vol 16 (4) ◽  
pp. 294-302 ◽  
Author(s):  
Shahid Akbar ◽  
Maqsood Hayat ◽  
Muhammad Kabir ◽  
Muhammad Iqbal
Keyword(s):  
Feature Extraction ◽  
Amino Acid ◽  
Antifreeze Proteins ◽  
Protein Sequences ◽  
Sampling Technique ◽  
Lower Class ◽  
Success Rates ◽  
Throughput Model ◽  
Extraction Scheme ◽  
Living Organisms

Antifreeze proteins (AFPs) perform distinguishable roles in maintaining homeostatic conditions of living organisms and protect their cell and body from freezing in extremely cold conditions. Owing to high diversity in protein sequences and structures, the discrimination of AFPs from non- AFPs through experimental approaches is expensive and lengthy. It is, therefore, vastly desirable to propose a computational intelligent and high throughput model that truly reflects AFPs quickly and accurately. In a sequel, a new predictor called “iAFP-gap-SMOTE” is proposed for the identification of AFPs. Protein sequences are expressed by adopting three numerical feature extraction schemes namely; Split Amino Acid Composition, G-gap di-peptide Composition and Reduce Amino Acid alphabet composition. Usually, classification hypothesis biased towards majority class in case of the imbalanced dataset. Oversampling technique Synthetic Minority Over-sampling Technique is employed in order to increase the instances of the lower class and control the biasness. 10-fold cross-validation test is applied to appraise the success rates of “iAFP-gap-SMOTE” model. After the empirical investigation, “iAFP-gap-SMOTE” model obtained 95.02% accuracy. The comparison suggested that the accuracy of” iAFP-gap-SMOTE” model is higher than that of the present techniques in the literature so far. It is greatly recommended that our proposed model “iAFP-gap-SMOTE” might be helpful for the research community and academia.

Sign in / Sign up

Export Citation Format

Share Document