Biology subjects

Frolova, D.

Publications and source records attributed to Frolova, D..

5 recordsLinked to original sources

bioRxiv2025

Bento: Benchmarking Classical and AI Docking on Drug Design-Relevant Data

AO_SCPLOWBSTRACTC_SCPLOWRecent advances in artificial intelligence have introduced deep learning and co-folding approaches for predicting protein-ligand complexes, raising the question of their applicability and how they compare with classical docking methods. In this work, we present a thorough benchmarking study of eleven tools for protein-ligand interaction prediction, spanning classical molecular docking methods, deep learning-based models, and co-folding algorithms. While most related benchmarking efforts primarily assess the generalization capacity, we extend the analysis to also evaluate the performance on drug design-relevant data and across different classes of protein-ligand complexes. Here, we introduce BO_SCPLOWENTOC_SCPLOW, a comprehensive benchmark that evaluates 11 tools for protein-ligand interaction prediction - both established and recently developed - across four test datasets and multiple derived subsets in a pocket-aware setup. We show that 1) careful dataset curation is essential - filtering by pocket structural similarity and controlling ligand complexity exposes generalization failures that are obscured in conventional benchmarks; 2) classical and deep learning-based docking tools perform similarly well on drug-like ligands, making them comparably useful for virtual screening, with physics-based methods offering a clear advantage in speed; 3) co-folding tools outperform other approaches on structurally complex ligands, whereas most methods achieve similar accuracy on regular small molecules; and 4) all methods struggle to generalize to unseen pockets, with deep learning models being the most prone to overfitting. Overall, our results show that while current docking and DL-based approaches are reliable for many drug-design-relevant scenarios, genuine pocket-level generalization remains an open challenge. BO_SCPLOWENTOC_SCPLOW provides a rigorous and transparent framework for diagnosing these limitations and guiding the development of more robust protein-ligand prediction models. The data and code of Bento are available at https://github.com/LigandPro/Bento.

bioinformatics↗

bioRxiv2025

Clustering of plasmid genomes for genomic epidemiology by using rearrangement distances, with pling

Integration of plasmids into genomic epidemiology is challenging, because there are no clearly defined evolving-units (equivalent to species), and because plasmids appear to evolve as much by structural change (rearrangements, insertions and deletions) as by mutation (1). Further, plasmids transfer horizontally between bacterial hosts (2), and thus a model beyond just a phylogeny is needed to integrate their genetic information with that of their hosts. Pling (3) is a tool designed to measure a genetic distance between plasmids that is related to how they empirically appear to evolve, by measuring the distance between two plasmids as the minimum number of structural changes needed to change one plasmid into the other (ignoring SNP differences). Having done this, it constructs a relatedness network of the plasmids under study, and then clusters them into groups that are credibly recently related. We give here a protocol for running pling, and how we integrate its information with plasmid typing and SNP information. Together, these provide a system for deciding which plasmids are worth treating as "the same plasmid" for the purposes of epidemiology, quantifying their relatedness in terms of rearrangements and SNPs, and then seeing how they are distributed across the host phylogeny.

bioinformatics↗

bioRxiv2024

TRACKING CLONAL AND PLASMID TRANSMISSION IN COLISTIN AND CARBAPENEM RESISTANT KLEBSIELLA PNEUMONIAE

The surveillance of mobile genetic elements facilitating the spread of antimicrobial resistance genes has been challenging. Here, we tracked both clonal and plasmid transmission in colistin- and carbapenem-resistant K. pneumoniae using short and long read sequencing technologies. We observed three clonal transmissions, all containing IncL plasmids and blaNDM-1, although not co-located on the same plasmid. One IncL-blaNDM-1 plasmid had been transferred between ST392 and ST15, and the promiscuous IncL-blaOXA-48 plasmid was likely shared between a singleton and a clonal transmission of ST392. Plasmids within clonal outbreaks and between clusters and STs had 0-2 SNP differences, showing high stability upon transfer to same or different STs. The simplest explanation of a single common IncL-blaNDM-1 plasmid spreading was in fact false, and we found blaNDM-1 in the context of five different plasmids, emphasizing the need to investigate plasmid-mediated transmission for effective containment of outbreaks. IMPORTANCEAntimicrobial resistance occupies a central stage in global public health emergencies. Recently, efforts to track the genetic elements that facilitate the spread of resistance genes to determine so-called plasmid outbreaks have been described, however, such short read sequencing hinders full knowledge about plasmid structure and makes this approach very challenging to implement. In this study we used both short and long read sequencing in clinical Klebsiella pneumoniae from University Hospital Centre Zagreb, Croatia which were resistant to both last resort antibiotics colistin and carbapenem. Our results show complex transmission networks and sharing of plasmids, emphasizing multiple transmissions of plasmids harbouring carbapenem and/or colistin resistance genes between and within K. pneumoniae clones. Only full length sequencing plus a novel way of determining plasmid clusters resulted in the full picture, showing how future active monitoring of plasmids as a vital tool for infection prevention and control could be implemented.

microbiology↗

bioRxiv2024

Applying rearrangement distances to enable plasmid epidemiology with pling

Plasmids are a key vector of antibiotic resistance, but the current bioinformatics toolkit is not well suited to tracking them. The rapid structural changes seen in plasmid genomes present considerable challenges to evolutionary and epidemiological analysis. Typical approaches are either low resolution (replicon typing) or use shared k-mer content to define a genetic distance. However this distance can both overestimate plasmid relatedness by ignoring rearrangements, and underestimate by over-penalising gene gain/loss. Therefore a model is needed which captures the key components of how plasmid genomes evolve structurally - through gene/block gain or loss, and rearrangement. A secondary requirement is to prevent promiscuous transposable elements (TEs) leading to over-clustering of unrelated plasmids. We choose the "Double Cut and Join Indel" model, in which plasmids are studied at a coarse level, as a sequence of signed integers (representing genes or aligned blocks), and the distance between two plasmids is the minimum number of rearrangement events or indels needed to transform one into the other. We show how this gives much more meaningful distances between plasmids. We introduce a software workflow pling (https://github.com/iqbal-lab-org/pling), which uses the DCJ-Indel model, to calculate distances between plasmids and then cluster them. In our approach, we combine containment distances and DCJ-Indel distances to build a TE-aware plasmid network. We demonstrate superior performance and interpretability to other plasmid clustering tools on the "Russian Doll" dataset and a hospital transmission dataset. Impact statementStudying plasmid transmission is a necessary component of understanding antibiotic resistance spread, but identifying recently related plasmids is difficult and often requires manual curation. Pling simplifies this by leveraging a combination of containment distances and rearrangement distances to cluster plasmids. The outcome are clusters of recently related plasmids with a clear backbone and relatively large core genomes, in contrast to other tools which sometimes overcluster. Additionally the network constructed by pling provides a framework with which to spot evolutionary events, such as potential fusions of plasmids and spread of transposable elements. Data summarySupplementary information and figures are available as an additional PDF. The tool presented in this paper is available under https://github.com/iqbal-lab-org/pling. Additional computational analysis and scripts are described and provided under https://github.com/babayagaofficial/pling_paper_analyses. The sequence data used can be found under BioProject no. PRJNA246471 in the National Center for Biotechnology Information for the "Russian doll" dataset (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA246471), and under Project no. PRJEB31034 in European Nucleotide Archive for the "Addenbrookes" dataset (https://www.ebi.ac.uk/ena/browser/view/PRJEB30134). All other genome sequences used were sourced from PLSDB (https://ccb-microbe.cs.uni-saarland.de/plsdb/), and lists of accession numbers can be found in the additional analysis github.

bioinformatics↗

bioRxiv2024

MULAN: Multimodal Protein Language Model for Sequence and Structure Encoding

Most protein language models (PLMs), which are used to produce high-quality protein representations, use only protein sequences during training. However, the known protein structure is crucial in many protein property prediction tasks, so there is a growing interest in incorporating the knowledge about the protein structure into a PLM. In this study, we propose MULAN, a MULtimodal PLM for both sequence and ANgle-based structure encoding. MULAN has a pre-trained sequence encoder and an introduced Structure Adapter, which are then fused and trained together. According to the evaluation on 7 downstream tasks of various nature, both small and medium-sized MULAN models show consistent improvement in quality compared to both sequence-only ESM-2 and structure-aware SaProt. Importantly, our model offers a cheap increase in the structural awareness of the protein representations due to finetuning of existing PLMs instead of training from scratch. We perform a detailed analysis of the proposed model and demonstrate its awareness of the protein structure. The implementation, training data and model checkpoints are available at https://github.com/DFrolova/MULAN.

bioinformatics↗