ATAndrey Tovchigrechko

Andrey Tovchigrechko

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

Andrey Tovchigrechko has a total of 41 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Assessing data size requirements for training generalizable sequence-based TCR specificity models via pan-allelic MHC-I non-self ligandome evaluation

    by Antoine Delaunay, Miles McGibbon, Bachir Djermani, Nikolai Gorbushin, Sergio Chaves Garcia-Mascaraque, Isaac Rayment, Ilya Kizhvatov, Cecile Petit, Maren Lang, Karim Beguir, et al.

    Quickly identifying which T cell receptors (TCRs) specifically bind patient-unique neoepitopes is a critical challenge for personalized TCR cell therapy in oncology. Due to enormous diversity of both TCR and neoepitope repertoires, a machine learning predictor of TCR-pMHC specificity for personalized therapy must generalize to TCRs and epitopes not seen in the training data. For the first time, we estimate the necessary size of such training data. We first show that published models fail to generalize beyond a single-residue dissimilarity to the epitope training set distribution. We then impute the possible mutated ligandome across the 34 most prevalent human MHC alleles and represent it as a graph based on our established dissimilarity cutoff. By finding the dominating set of this graph, we estimate that between one and 100 million epitopes are required to train a generalizable sequence-based TCR specificity prediction model - 1000 times the size of current public data.

  • Open Access

    FrameDiPT: SE(3) Diffusion Model for Protein Structure Inpainting

    by Cheng Zhang, Adam Leach, Thomas Makkink, Miguel Arbesú, Ibtissem Kadri, Daniel Luo, Liron Mizrahi, Sabrine Krichen, Maren Lang, Andrey Tovchigrechko, et al.

    Protein structure prediction field has been revolutionised by deep learning with protein folding models such as AlphaFold 2 and ESMFold. These models enable rapid in silico prediction and have been integrated into de novo protein design and protein-protein interaction (PPI) prediction. However, biologically relevant features dependent on conformational distributions cannot be estimated with these models. Diffusion models, a novel class of generative models, have been developed to learn conformational distributions and applied to de novo protein design. Limited work has been done on protein structure inpainting, where a masked section is recovered by simultaneously conditioning on its sequence and the rest of the structure. In this work, we propose FrameD iff i n P ain T ing (FrameDiPT), a generalised model for protein inpainting. This is important for T-cells given the hyper-variability of the complementarity determining region (CDR) loops. We evaluated the model on CDR loop design for T-cell receptors and achieved comparable prediction accuracy to ProteinGenerator and RFdiffusion with limited training data and learnable parameters. Different from deterministic structure prediction models, FrameDiPT captures the conformational distribution at different regions and binding states, highlighting a key advantage of generative models. The model and inference code have been released 1 .


  • Open Access

    Haemagglutinin mutations and glycosylation changes shaped the 2012/13 influenza A(H3N2) epidemic, Houston, Texas

    by K M Stucker, S A Schobel, R J Olsen, H L Hodges, X Lin, R A Halpin, N Fedorova, T B Stockwell, A Tovchigrechko, S R Das, et al.

    While the early start and higher intensity of the 2012/13 influenza A virus (IAV) epidemic was not unprecedented, it was the first IAV epidemic season since the 2009 H1N1 influenza pandemic where the H3N2 subtype predominated. We directly sequenced the genomes of 154 H3N2 clinical specimens collected throughout the epidemic to better understand the evolution of H3N2 strains and to inform the H3N2 vaccine selection process. Phylogenetic analyses indicated that multiple co-circulating clades and continual antigenic drift in the haemagglutinin (HA) of clades 5, 3A, and 3C, with the evolution of a new 3C subgroup (3C-2012/13), were the driving causes of the epidemic. Drift variants contained HA substitutions and alterations in the potential N-linked glycosylation sites of HA. Antigenic analysis demonstrated that viruses in the emerging subclade 3C.3 and subgroup 3C-2012/13 were not well inhibited by antisera generated against the 3C.1 vaccine strains used for the 2012/13 (A/Victoria/361/2011) or 2013/14 (A/Texas/50/2012) seasons. Our data support updating the H3N2 vaccine strain to a clade 3C.2 or 3C.3-like strain or a subclade that has drifted further. They also underscore the challenges in vaccine strain selection, particularly regarding HA and neuraminidase substitutions derived during laboratory passage that may alter antigenic testing accuracy.


  • The size of the intermolecular energy funnel in protein-protein interactions

    by Jagtar Hunjan, Andrey Tovchigrechko, Ying Gao, Ilya A. Vakser

    Revealing the fundamental principles of protein interactions is essential for the basic knowledge of molecular processes and designing better predictive tools. Protein docking procedures allow systematic sampling of intermolecular energy landscapes, revealing the distribution of energy basins and their characteristics. A systematic search docking procedure GRAMM-X was applied to a comprehensive nonredundant database of nonobligate protein–protein complexes to determine the size of the intermolecular energy funnel. The unbound structures were simulated using rotamer library. The procedure generated grid-based matches, based on a smoothed Lennard-Jones potential, and minimized them off the grid with the same potential. The minimization generated a distribution of distances, based on a variety of metrics, between the grid-based and the minimized matches. The metric selected for the analysis, ligand interface RMSD, provided three independent estimates of the funnel size: based on the distribution amplitude for the near-native matches, deviation from random, and correlation with the energy values. The three methods converge to similar estimates of ∼6–8 Å ligand interface RMSD. The results indicated dependence of the funnel size on the type of the complex (smaller for antigen–antibody, medium for enzyme–inhibitor, and larger for the rest of the complexes) and the funnel size correlation with the size of the interface. Guidelines for the optimal sampling of docking coordinates, based on the funnel size estimates, were explored. Proteins 2008.