PowerNovo: de novo peptide sequencing via tandem mass spectrometry using an ensemble of transformer and BERT models.

The primary objective of analyzing the data obtained in a mass spectrometry-based proteomic experiment is peptide and protein identification, or correct assignment of the tandem mass spectrum to one amino acid sequence. Comparison of empirical fragment spectra with the theoretical predicted one or matching with the collected spectra library are commonly accepted strategies of proteins identification and defining of their amino acid sequences. Although these approaches are widely used and are appreciably efficient for the well-characterized model organisms or measured proteins, they cannot detect novel peptide sequences that have not been previously annotated or are rare. This study presents PowerNovo tool for de novo sequencing of proteins using tandem mass spectra acquired in a variety of types of mass analyzers and different fragmentation techniques. PowerNovo involves an ensemble of models for peptide sequencing: model for detecting regularities in tandem mass spectra, precursors, and fragment ions and a natural language processing model, which has a function of peptide sequence quality assessment and helps with reconstruction of noisy sequences. The results of testing showed that the performance of PowerNovo is comparable and even better than widely utilized PointNovo, DeepNovo, Casanovo, and Novor packages. Also, PowerNovo provides complete cycle of processing (pipeline) of mass spectrometry data and, along with predicting the peptide sequence, involves the peptide assembly and protein inference blocks.

SAFoldNet: A Novel Tool for Discovering and Aligning Three-Dimensional Protein Structures Based on a Neural Network.

The development and improvement of methods for comparing and searching for three-dimensional protein structures remain urgent tasks in modern structural biology. To solve this problem, we developed a new tool, SAFoldNet, which allows for searching, aligning, superimposing, and determining the exact coordinates of fragments of protein structures. The proposed search and alignment tool was built using neural networking. Specifically, we implemented the integrative synergy of neural network predictions and the well-known BLAST algorithm for searching and aligning sequences. The proposed method involves multistage processing, comprising a stage for converting the geometry of protein structures into sequences of a structural alphabet using a neural network, a search stage for forming a set of candidate structures, and a refinement stage for calculating the structural alignment and overlap and evaluating the similarity with the starting structure of the search. The effectiveness and practical applicability of the proposed tool were compared with those of several widely used services for searching and aligning protein structures. The results of the comparisons confirmed that the proposed method is effective and competitive relative to the available modern services. Furthermore, using the proposed approach, a service with a user-friendly web interface was developed, which allows for searching, aligning, and superimposing protein structures; determining the location of protein fragments; mapping onto a protein molecule chain; and providing structural similarity metrices (expected value and root mean square deviation).

Modeling Side Chains in the Three-Dimensional Structure of Proteins for Post-Translational Modifications.

Amino acid substitutions and post-translational modifications (PTMs) play a crucial role in many cellular processes by directly affecting the structural and dynamic features of protein interaction. Despite their importance, the understanding of protein PTMs at the structural level is still largely incomplete. The Protein Data Bank contains a relatively small number of 3D structures having post-translational modifications. Although recent years have witnessed significant progress in three-dimensional modeling (3D) of proteins using neural networks, the problem related to predicting accurate PTMs in proteins has been largely ignored. Predicting accurate 3D PTM models in proteins is closely related to another fundamental problem: predicting the correct side-chain conformations of amino acid residues in proteins. An analysis of publications as well as the paid and free software packages for modeling three-dimensional structures showed that most of them focus on working with unmodified proteins and canonical amino acid residues; the number of articles and software packages placing emphasis on modeling three-dimensional PTM structures is an order of magnitude smaller. This paper focuses on modeling the side-chain conformations of proteins containing PTMs (nonstandard amino acid residues). We collected our own libraries comprising the most frequently observed PTMs from the PDB and implemented a number of algorithms for predicting the side-chain conformation at modification points and in the immediate environment of the protein. A comprehensive analysis of both the algorithms per se and compared to the common Rosetta and FoldX structure modeling packages was also carried out. The proposed algorithmic solutions are comparable in their characteristics to the well-known Rosetta and FoldX packages for the modeling of three-dimensional structures and have great potential for further development and optimization. The source code of algorithmic solutions has been deposited to and is available at the GitHub source.

Consolidation of metabolomic, proteomic, and GWAS data in connective model of schizophrenia.

Despite of multiple systematic studies of schizophrenia based on proteomics, metabolomics, and genome-wide significant loci, reconstruction of underlying mechanism is still a challenging task. Combination of the advanced data for quantitative proteomics, metabolomics, and genome-wide association study (GWAS) can enhance the current fundamental knowledge about molecular pathogenesis of schizophrenia. In this study, we utilized quantitative proteomic and metabolomic assay, and high throughput genotyping for the GWAS study. We identified 20 differently expressed proteins that were validated on an independent cohort of patients with schizophrenia, including ALS, A1AG1, PEDF, VTDB, CERU, APOB, APOH, FASN, GPX3, etc. and almost half of them are new for schizophrenia. The metabolomic survey revealed 18 group-specific compounds, most of which were the part of transformation of tyrosine and steroids with the prevalence to androgens (androsterone sulfate, thyroliberin, thyroxine, dihydrotestosterone, androstenedione, cholesterol sulfate, metanephrine, dopaquinone, etc.). The GWAS assay mostly failed to reveal significantly associated loci therefore 52 loci with the smoothened p < 10-5 were fractionally integrated into proteome-metabolome data. We integrated three omics layers and powered them by the quantitative analysis to propose a map of molecular events associated with schizophrenia psychopathology. The resulting interplay between different molecular layers emphasizes a strict implication of lipids transport, oxidative stress, imbalance in steroidogenesis and associated impartments of thyroid hormones as key interconnected nodes essential for understanding of how the regulation of distinct metabolic axis is achieved and what happens in the conditioned proteome and metabolome to produce a schizophrenia-specific pattern.

Pilot data of serum proteins from children with autism spectrum disorders.

Protein profiles of 13 serum samples from children with autism spectrum disorders (ASD) and 11 serum samples from healthy volunteers was obtained using panoramic ultra-high resolution mass spectrometry. The analysis of measurements was performed using the proteomics search engine. We identified a group of 74 proteins which we term a "protein fingerprint" specific for serum samples collected from children with autism. Components of the protein fingerprint are involved in hemostasis maintenance including biological regulation, the response to stimulus, regulation of metabolism, and proteins of the immune system.