ASAnatoly Shalyto

Anatoly Shalyto

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Anatoly Shalyto has a total of 145 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Incorporating negative examples into Hidden Markov Model-based classification of peptide sequences

    by V. A. Polezhaeva, D. A. Kleverov, A. A. Shalyto, M. Artyomov, М. Артемов

    Hidden Markov Models (HMMs) trained to identify binding regions in peptide sequences have demonstrated the ability to uncover shared amino acid patterns in peptides bound to major histocompatibility complex molecules. In this work, we present an enhanced approach for predicting peptide binding using an ensemble of HMMs. Building on a previously proposed method, we extend it to a classification setting by incorporating both binding (positive) and non-binding (negative) peptide sequences. Our strategy involves training two sets of models on these distinct datasets and selecting ensemble members based on conditional probability estimates. The method was evaluated across six alleles of major histocompatibility complex using two model architectures: simplified architecture with 9 states representing the peptide binding core region and two cycle-states for the amino acids outside this region, and extended architecture, in which each cycle state was replaced by 9 additional states. Models evaluated in comparison with the state-of-the-art MixMHC2pred predictor. Results show a statistically significant improvement in prediction accuracy. Notably, incorporating non-binding peptides during training improved performance in several cases, highlighting the importance of background sequence information in distinguishing binding-specific patterns.

  • Open Access

    Comparative analysis method for time series data objects represented as sets of strings based on de Bruijn graphs

    by A. B. Ivanov, A. A. Shalyto, V. I. Ulyantsev

    The paper considers the comparative analysis of string datasets represented as time series of samples. We propose a method to increase the accuracy of determining differences between two samples. Based on this method, a method for analyzing time series of three samples has been developed, which allows for more accurate changes investigation between samples. The use of three samples in the analysis is due to the specific nature of the practical task of processing metagenomic sample sequencing data, obtaining a larger number of which is very resource-intensive. To classify strings from one sample into detected and undetected strings in another sample, a method of comparing two samples using k-mers and the de Bruijn graph is proposed. It implements decision rules based on statistics of the frequency of k-mers occurrence, different values of the parameter k, and information about possible errors in the strings. To analyze time series of three samples (the original and final sample for one object and the modifying sample for another object), a method based on pairwise comparison of samples is developed. It is used to divide the strings of each sample into groups depending on the detection of strings in other samples. The developed method for analyzing time series has been tested on two types of generated metagenomic data, represented as a set of strings. It was shown that the method allows distinguishing organisms that have differences in genomes in at least one symbol for every 10,000 symbols. High (more than 80 %) recall and precision of the results of string classification were demonstrated when analyzing simulated complex data, with properties comparable to real data. The developed method allows comparing metagenomic samples represented as a set of strings using only the data itself and without requiring additional information. This allows for a more accurate analysis compared to existing methods that compare samples based on the results of string classification in taxonomic annotation databases. The developed methods can also be used in other areas of string data processing such as analyzing changes in author style when writing a series of texts.

  • Open Access

    Method for identifying the active module in biological graphs with multi-component vertex weights

    by D. A. Usoltsev, I. I. Molotkov, M. N. Artomov, A. A. Sergushichev, A. A. Shalyto, Н. Н. Артемов

    An active module in biological graphs is a connected subgraph whose vertices share a common biological function. To identify an active module, one must first construct a weighted biological graph. The weight of each vertex is calculated based on biological experiments investigating the target biological function. However, the results of a single experiment may not fully describe the desired active module, covering only part of it and potentially introducing uncertainty into the vertex weights. This work demonstrates that employing Fisher’s method to integrate data from multiple experiments followed by applying a Markov chain Monte Carlo (MCMC) and machine learning–based approach to the results of Fisher’s method, enables more effective identification of active modules in biological graphs. The study utilizes the InWebIM protein–protein interaction graph, a human brain reconstruction graph from the BigBrain project, and a gene graph for the organism Caenorhabditis elegans. To combine the results of several experiments into a single outcome within one graph, Fisher’s method is applied. Afterwards, the search for active modules is conducted using an MCMC and machine learning-based method. To validate the proposed method on real data, results from GenomeWide Association Studies on schizophrenia and smoking are used, along with the gene expression matrix of patients with skin melanoma from the TCGA project. Applying Fisher’s method makes it possible to consider the results of multiple biological experiments simultaneously. Subsequent use of the MCMC and machine learning–based method improves the accuracy of identifying active modules compared to ranking graph vertices solely by Fisher’s method. Considering the results of multiple biological experiments when determining active modules plays a crucial role in increasing the accuracy of identifying the vertices of the active module. This, in turn, promotes a deeper understanding of the biological mechanisms of diseases, which can be of great significance for the development of new diagnostic and therapeutic methods.

  • Open Access

    Feature extraction methods for metagenome de Bruijn graphs collections based on samples classification information

    by A. B. Ivanov, A. A. Shalyto, V. I. Ulyantsev

    The paper considers the comparative analysis of metagenomic samples collections using de Bruijn graphs. We propose methods for automatic feature extraction based on the results of comparative sample analysis, expert metadata, and statistical tests to improve the accuracy of classification models. In this paper features are connected subgraphs of the de Bruijn graph. The first method, named unique_kmers, is used to extract strings of length k (k-mers) that occur only in samples of the certain class. The second method, named stats_kmers, is used to extract k-mers whose frequency of occurrence statistically differs between sample classes. To extract interpretable features, a third method has been developed that implements the extraction of subgraphs from de Bruijn graphs based on the selected nodes obtained as a result of applying one of the first two methods. Data analysis consists of two stages: firstly, unique_kmers or stats_kmers method is applied for data preprocessing, secondly, the third method is applied to obtain interpretable features. The methods were tested on four generated datasets that model the properties of real metagenomic communities such as the presence of similar species (strains) or differences in the relative abundance of bacteria. The developed methods were used to extract features. Machine learning model was trained in extracted features to classify samples from the test datasets. For comparison, the results of taxonomic annotation of samples using the Kraken2 program were used as features. It was shown that the accuracy of samples classification increased when using features obtained using the proposed methods in classification models compared to classification models trained on taxonomic features. The developed methods are useful for comparative analysis of metagenomic sequencing data and can form the basis of decision support systems, for example, in human diseases diagnostics based on gut microbiota sequencing data.

  • Open Access

    Surface defect detection with limited data based on SSD detector and Siamese networks

    by N.P. Novgorodcev, K.A. Baturina, V.A. Efimova, A.A. Shalyto

    This study presents an algorithm for the problem of detecting defects on hard surfaces when trained with zero or a small number of examples, addressing the challenge of limited data availability. The existing defect detection methodology using machine vision is enhanced. A hybrid approach is proposed, combining the strengths of the SSD detector and Siamese Neural Networks (SNN). The SSD detector extracts feature vector representations from images, while the SNNs are used to construct the feature space. The new approach demonstrates high accuracy in detecting both known and previously unseen defects in the training dataset. Based on testing across seven different datasets, the model showed good performance in scenarios with a limited number of training examples. A comparative analysis with existing models highlights the high performance of the proposed approach and its potential as an innovative and effective solution for the universal detection of defects on hard surfaces.


  • Open Access

    Predicting gene-disease associations using a heterogeneous graph neural network

    by D.A. Sidorenko, Shalyto Anatoly A. Sidorenko Denis A.

    The research presents the development of a heterogeneous graph neural network model for predicting gene-disease using existing genomic and medical data. The novelty of the approach is in integrating the principles of graph neural networks and heterogeneous information networks for efficient processing of structured data and consideration of complex genepathology interactions. The solution proposed is a heterogeneous graph neural network which utilizes a heterogeneous graph structure for representing genes, diseases, and their relationships. The performance of the developed model was evaluated on the DisGeNET, LASTFM, YELP datasets. On these datasets, a comparison was made with current SOTA models. The comparison results demonstrated that the proposed model outperforms other models in terms of Average Precision (AP), F1-measure (F1@S), Hit@k, Area Under Receiver Operating Characteristic curve (AUROC) in predicting “gene-disease” associations. The model developed serves as a tool for bioinformatics analysis and can aid researchers and doctors in studying genetic diseases. This could expedite the discovery of new drug targets and the advancement of personalized medicine.

  • Open Access

    A method for constructing interpretable hidden Markov models for the task of identifying binding cores in sequences

    by D.A. Kleverov, A.A. Shalyto, Maxim Artyomov

    Solving the problem of predicting the immune response against foreign protein sequence fragments processed by cells is one of the major milestones on the road to the personalized cancer vaccine development. The selection of peptides participating in the immune response is a complex multi-stage process of filtering initial sequences to present their fragments on the cell surface. The most studied task regarding this filtering nowadays is the prediction of the binding probability of peptides to major histocompatibility complex molecules. Modern methods for predicting this stage are usually based on algorithms using artificial neural networks, which make it impossible to interpret the result predictions of such models. One of the methods to overcome this limitation is the use of interpretable hidden Markov models. In this work, an analysis of the binding prediction task is performed. As a result, a method for constructing interpretable models that consider domain-specific constraints and requirements is proposed. A method for the constriction, training and interpretation of hidden Markov models was proposed for each class of molecules. The construction and training are based on maintaining the model architecture capable of extracting and visualizing the binding core of the peptide. Interpretation is possible through the analysis of the model graph. The proposed method is tested in the task of training a model that not only enables prediction but also facilitates determining the position of the peptide binding core and the distribution of amino acids within the core. Prediction models were trained for two types of molecules using binding data. The distributions of amino acids in the binding core match the state distributions of the model. Sequence patterns of such regions extracted using the trained models for two sets of peptide data correspond to patterns from public databases, confirming the successful validation of the method. Interpretable models provide a better description of the problem domain and help to draw a conclusion about peptide characteristics based on information extracted from the model. This information will allow researchers to better understand other steps of peptide processing involved in the immune response. For example, one can study relationships between these steps or perform a transfer of knowledge from models trained for one step to others. Using this knowledge will allow the training of the models under conditions of limited training data.


  • Open Access

    Validation of state machine specifications

    by A.A. Shalyto

    The paper presents a new view on the quality assurance of state machine programs. Instead of the term “verification of state machine programs”, it is proposed to use the terms “verification of state machine models” and “validation of state machine specifications”. The first of which is applicable in the presence of a formal specification, and the second — in its absence, which is more typical for practice. This allows a more meaningful approach to understanding how to ensure the quality of state machine programs.


  • Continuous control algorithms for conveyer belt routing based on multi-agent deep reinforcement learning

    by Yaroslav Zhurba, Andrei Filchenkov, Arthur Azarov, Anatoly Shalyto

    Introduction: We consider the problem of routing of piece cargo by a conveyor system. When moving cargo pieces, it is necessary not only to minimize the time of transportation, but also to minimize the energy spent on it. Purpose: Development of a routing algorithm that is adaptive to changes in the topology of the routing graph and is able to optimize the delivery time and the consumed energy. Results: We propose an algorithm based on multi-agent deep reinforcement learning that places agents at the vertices of a conveyor network graph and uses a new state value function. The algorithm has two tunable parameters: the length of the path along which the state value function is calculated, and the learning coefficient. Through the selection of parameters, we have revealed that the optimal values are 2 and 1, respectively. An experimental study of the algorithm using a simulation model has shown that it allows to reduce the number of collisions of moving objects to zero, demonstrates stable results for both optimized scores, and also leads to a lower energy consumption compared with the method used as a baseline. Practical relevance: The proposed algorithm can be used to reduce delivery time and energy when managing conveyor systems.

  • Open Access

    Multi-agent adaptive routing by multi-head-attention-based twin agents using reinforcement learning

    by Timofey A. Gribanov, Andrey A. Filchenkov, Artur A. Azarov, A.A. Shalyto

    A regular condition, typical for packet routing, for the problem of cargo transportation, and for the problem of flow control, is the variability of the graph. Reinforcement learning based adaptive routing algorithms are designed to solve the routing problem with this condition. However, with significant changes in the graph, the existing routing algorithms require complete retraining. To handle this challenge, we propose a novel method based on multi-agent modeling with twin-agents for which new neural network architecture with multi-headed internal attention is proposed, pre-trained within the framework of the multi-view learning paradigm. An agent in such a paradigm uses a vertex as an input, twins of the main agent are placed at the vertices of the graph and select a neighbor to which the object should be transferred. We carried out a comparative analysis with the existing DQN-LE-routing multi-agent routing algorithm on two stages: pre-training and simulation. In both cases, launches were considered by changing the topology during testing or simulation. Experiments have shown that the proposed adaptability enhancement method provides global adaptability by increasing delivery time only by 14.5 % after global changes occur. The proposed method can be used to solve routing problems with complex path evaluation functions and dynamically changing graph topologies, for example, in transport logistics and for managing conveyor belts in production.