Biology subjects

Simecek, P.

Publications and source records attributed to Simecek, P..

12 recordsLinked to original sources

bioRxiv2026

DNA Compression with Genomic Language Models: Tokenization, Benchmarking, and an Information-Content Map

Lossless compression and probabilistic sequence modeling are two faces of the same coin: a model that assigns high probability to a sequence can encode it in few bits via arithmetic coding. We exploit this duality to evaluate genomic language models as compressors of DNA, using compression primarily as an objective probe of generative sequence modeling rather than as a deployable storage system. We release DNAGPT2, a family of ten GPT-2-small models pretrained for one epoch on a single A40 using the DNABERT2 multi-species corpus that differ only in byte-pair encoding vocabulary size. Coupled with arithmetic coding, the best model reaches 1.47 bits per base (bpb) on the T2T human genome, fourth in the Cobilab compression benchmark and ahead of every general-purpose compressor. Our results suggest that NLP-style tokenization choices may be suboptimal for DNA: a 32-token BPE vocabulary compresses better than larger vocabularies. We also find that, in this benchmark, published long-context genomic LMs underperform a much shorter-context BPE GPT-2; we discuss in Section 5 that this is not a controlled context-length ablation, since the compared models also differ in architecture, training data, parameter count, and tokenization. Finally, we compute a per-nucleotide information-content map of the human genome and show that exons, introns, intergenic regions, and Alu repeats have statistically distinct information profiles.

bioinformatics↗

bioRxiv2026

Advancing Knotted Protein Design with ESM3: Guided Generation and Topological Insights

Multimodal protein language models have transformed protein design, yet their capacity to capture complex topological features remains poorly understood. We use knotted proteins, rare structures in which the backbone forms a nontrivial topological knot, as a test case to probe this capacity using ESM3, a generative protein language model. Topology-aware guided decoding strongly enriches ESM3 outputs for knotted topologies, producing structures classified as knotted at an 89% success rate (95% CI: 81- 94%), compared to ~0.5% for unguided diffusion-based approaches. A confidence analysis shows that freshly generated artificial knots have lower ESM3 pLDDT and pTM than real knotted proteins evaluated under the same pipeline, motivating a cautious interpretation of generated examples as model samples pending independent validation. In contrast, the robustness analyses on real knotted proteins are high-confidence: on average 84% of the protein sequence must be altered before the knot breaks, and the loss follows a sharp threshold rather than gradual degradation. Strikingly, structural drift accumulates well before topological disruption, suggesting that topology is more robust than specific three-dimensional arrangement. These findings position knotted proteins as a useful probe of how generative protein models represent rare, global structural features.

bioengineering↗

bioRxiv2025

Balancing Speed and Precision in Protein Folding: A Comparison of AlphaFold2, ESMFold, and OmegaFold

We compared the performance of three widely used protein structure prediction tools--AlphaFold2, ESMFold, and OmegaFold--using a dataset of over 1,300 newly created records from the PDB database. These structures, resolved between July 2022 and July 2024, ensure unbiased evaluation, as they were unavailable during the training of these tools. Using metrics such as root mean square deviation (RMSD), template modeling score (TM-score), and predicted local distance difference test (pLDDT), we found that AlphaFold2 consistently achieves the highest accuracy but depends on high-quality sequence alignments. In contrast, ESMFold and OmegaFold provide faster predictions and excel in challenging cases, such as rapidly evolving or designed proteins with limited sequence homology. Comparing ESMFold and OmegaFold, ESMFold achieves higher confidence scores (pLDDT) and structural similarity (TM-score). OmegaFold is competitive in specific contexts, such as de novo-designed proteins or sequences with limited evolutionary information. Additionally, we demonstrate that machine learning models trained on protein language model embeddings and pLDDT confidence scores can predict potential structure prediction failures, helping to identify challenging cases early in the pipeline.

bioinformatics↗

bioRxiv2023

Knot or Not? Sequence-Based Identification of Knotted Proteins With Machine Learning

Knotted proteins, although scarce, are crucial structural components of certain protein families, and their roles remain a topic of intense research. Capitalizing on the vast collection of protein structure predictions offered by AlphaFold, this study computationally examines the entire UniProt database to create a robust dataset of knotted and unknotted proteins. Utilizing this dataset, we develop a machine learning model capable of accurately predicting the presence of knots in protein structures solely from their amino acid sequences, with our best-performing model demonstrating a 98.5% overall accuracy. Unveiling the sequence factors that contribute to knot formation, we discover that proteins predicted to be unknotted from known knotted families are typically non-functional fragments missing a significant portion of the knot core. The study further explores the significance of the substrate binding site in knot formation, particularly within the SPOUT protein family. Our findings spotlight the potential of machine learning in enhancing our understanding of protein topology and propose further investigation into the role of knotted structures across other protein families. TOC Graphic O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=111 SRC="FIGDIR/small/556468v1_ufig1.gif" ALT="Figure 1"> View larger version (27K): org.highwire.dtl.DTLVardef@1466077org.highwire.dtl.DTLVardef@1674829org.highwire.dtl.DTLVardef@1b283a5org.highwire.dtl.DTLVardef@e0e962_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2022

Genomic Benchmarks: A Collection of Datasets for Genomic Sequence Classification

In this paper, we propose a collection of curated and easily accessible sequence classification datasets in the field of genomics. The proposed collection is based on a combination of novel datasets constructed from the mining of publicly available databases and existing datasets obtained from published articles. The main aim of this effort is to create a repository for shared datasets that will make machine learning for genomics more comparable and reproducible while reducing the over-head of researchers that want to enter the field. The collection currently contains eight datasets that focus on regulatory elements (promoters, enhancers, open chromatin region) from three model organisms: human, mouse, and roundworm. A simple convolution neural network is also included in a repository and can be used as a baseline model. Benchmarks and the baseline model are distributed as the Python package genomic-benchmarks, and the code is available at https://github.com/ML-Bioinfo-CEITEC/genomic_benchmarks.

bioinformatics↗

bioRxiv2022

Fine-Tuning Transformers For Genomic Tasks

Transformers are a type of neural network architecture that has been successfully used to achieve state-of-the-art performance in numerous natural language processing tasks. However, what about DNA, the language life written in the four-letter alphabet? In this paper, we review the current state of Transformers usage in genomics and molecular biology in general, introduce a collection of benchmark datasets for the classification of genomic sequences, and compare the performance of several model architectures on those benchmarks, including a BERT-like model for DNA sequences DNABERT as implemented in HuggingFace (armheb/DNA_bert_6 model). In particular, we explore the effect of pre-training on a large DNA corpus vs training from scratch (with randomized weights). The results presented here can be used for identification of functional elements in human and other genomes.

bioinformatics↗

bioRxiv2020

PENGUINN: Precise Exploration of Nuclear G-quadruplexes Using Interpretable Neural Networks

G-quadruplexes (G4s) are a class of stable structural nucleic acid motifs that are known to play a role in a wide spectrum of genomic functions, such as DNA replication and transcription. The classical understanding of G4 structure points to four variable length guanine strands joined by variable length stretches of other nucleotides. Experiments using G4 immunoprecipitation and sequencing experiments have produced a high number of highly probable G4 forming genomic sequences. The expense and technical difficulty of experimental techniques highlights the need for computational approaches of G4 identification. Here, we present PENGUINN, a machine learning method based on Convolutional Neural Networks, that learns the characteristics of G4 sequences and accurately predicts G4s outperforming the state-of-the-art. We provide both a standalone implementation of the trained model, and a web application that can be used to evaluate sequences for their G4 potential.

bioinformatics↗

bioRxiv2019

Chromosome-wide distribution and characterization of intersubspecific meiotic noncrossovers in mice

During meiosis, the recombination-initiating DNA double-strand breaks (DSBs) are repaired by crossovers or noncrossovers (gene conversions). While crossovers are easily detectable, the noncrossover identification is hampered by the small size of their converted tracts and the necessity of sequence polymorphism to detect them. We report identification and characterization of a mouse chromosome-wide set of noncrossovers by NGS sequencing of 10 mouse chromosome substitution strains. Based on 94 identified noncrossovers we determined the mean length of a conversion tract to be 32 basepairs. The spatial chromosome-wide distribution of noncrossovers and crossovers significantly differed, though both sets overlapped the known hotspots of PRDM9-directed histone methylation and DNA DSBs, thus proving their origin in the standard DSB repair pathway. A significant deficit of noncrossovers descending from asymmetric DSBs pointed to sister chromatids as an alternative template for their repair. The finding has implications for the molecular mechanism of hybrid sterility in mice.

genetics↗

bioRxiv2019

High-resolution QTL mapping with Diversity Outbred mice identifies genetic variants that impact gut microbiome composition

The composition of the gut microbiome is impacted by a complex array of factors, from nutrient composition and availability, to physical factors like temperature, pH, and flow rate, as well as interactions among the members of the microbial community. Many of these factors are affected by the host, raising the question of how host genetic variation impacts microbiome composition. Though human studies confirm this type of role for host genetics, its overall importance is still a subject of debate and remains difficult to study. The mouse model, by allowing the strict control of genetics, nutrition, and other environmental factors, has provided an excellent opportunity to extend this work, and the Diversity Outbred (DO) mice in particular present a chance to pinpoint host genetic variants that influence microbiome composition at different levels of generality. Here, we apply 16S rRNA gene sequencing to fecal samples of 247 DO male mice to estimate heritability and perform taxon-specific QTL mapping of microbial relative abundances revealing an increasingly heterogeneous picture of host function and microbial taxa at the host-microbiome interface. We present the first report of significant heritability of phylum Tenericutes in mice, and find novel QTL-spanning genes involved in antibacterial pathways, immune and inflammatory disease, and lipid metabolism.

genetics↗

bioRxiv2018

R/qtl2: software for mapping quantitative trait loci with high dimensional data and multi-parent populations

R/qtl2 is an interactive software environment for mapping quantitative trait loci (QTL) in experimental populations. The R/qtl2 software expands the scope of the widely-used R/qtl software package to include multi-parent populations derived from more than two founder strains, such as the Collaborative Cross and Diversity Outbred mice, heterogeneous stocks, and MAGIC plant populations. R/qtl2 is designed to handle modern high-density genotyping data and high-dimensional molecular phenotypes including gene expression and proteomics. R/qtl2 includes the ability to perform genome scans using a linear mixed model to account for population structure, and also includes features to impute SNPs based on founder strain genomes and to carry out association mapping. The R/qtl2 software provides all of the basic features needed for QTL mapping, including graphical displays and summary reports, and it can be extended through the creation of add-on packages. R/qtl2 comes with a test framework and is free and open source software written in the R and C++ programming languages.

genetics↗

bioRxiv2017

High-Resolution Maps of Mouse Reference Populations

Genetic reference panels are widely used to map complex, quantitative traits in model organisms. We have generated new high-resolution genetic maps of 259 mouse inbred strains from recombinant inbred strain panels (C57BL/6J x DBA/2J, ILS/IbgTejJ x ISS/IbgTejJ, C57BL/6J x A/J) and chromosome substitution strain panels (C57BL/6J-Chr# , C57BL/6J-Chr# , C57BL/6JChr# ). We genotyped all samples using the Affymetrix Mouse Diversity Array with an average inter-marker spacing of 4.3kb. The new genetic maps provide increased precision in the localization of recombination breakpoints compared to the previous maps. Although the strains were presumed to be fully inbred, we found residual heterozygosity in 40% of individual mice from five of the six panels. We also identified de novo deletions and duplications, in homozygous or heterozygous state, ranging in size from 21kb to 8.4Mb. Almost two-thirds (46 out of 76) of these deletions overlap exons of protein coding genes and may have phenotypic consequences. Twenty-nine putative gene conversions were identified in the chromosome substitution strains. We find that gene conversions are more likely to occur in regions where the homologous chromosomes are more similar. The raw genotyping data and genetic maps of these strain panels are available at http://churchill-lab.jax.org/website/MDA.

genetics↗

bioRxiv2017

The Effects of Sex and Diet on Physiology and Liver Gene Expression in Diversity Outbred Mice

Inter-individual variation in metabolic health and adiposity is driven by many factors. Diet composition and genetic background and the interactions between these two factors affect adiposity and related traits such as circulating cholesterol levels. In this study, we fed 850 Diversity Outbred mice, half females and half males, with either a standard chow diet or a high fat, high sucrose diet beginning at weaning and aged them to 26 weeks. We measured clinical chemistry and body composition at early and late time points during the study, and liver transcription at euthanasia. Males weighed more than females and mice on a high fat diet generally weighed more than those on chow. Many traits showed sex- or diet-specific changes as well as more complex sex by diet interactions. We mapped both the physiological and molecular traits and found that the genetic architecture of the physiological traits is complex, with many single locus associations potentially being driven by more than one polymorphism. For liver transcription, we find that local polymorphisms affect constitutive and sex-specific transcription, but that the response to diet is not affected by local polymorphisms. We identified two loci for circulating cholesterol levels. We performed mediation analysis by mapping the physiological traits, given liver transcript abundance and propose several genes that may be modifiers of the physiological traits. By including both physiological and molecular traits in our analyses, we have created deeper phenotypic profiles to identify additional significant contributors to complex metabolic outcomes such as polygenic obesity. We make the phenotype, liver transcript and genotype data publicly available as a resource for the research community.

genetics↗