Biology subjects

Kulakovskiy, I. V.

Publications and source records attributed to Kulakovskiy, I. V..

At least 19 recordsLinked to original sources

bioRxiv2026

EPIC: An open community challenge for sequence-based prediction of transcription initiation in five non-model metazoans

Predicting gene expression from DNA sequence is a central problem with critical biological and clinical implications. Recent sequence-to-function models are reported to achieve improved performance, yet it remains unclear how much of what they learn reflects genuine principles of eukaryotic transcription initiation and to what extent they are able to generalize beyond humans and other primary model species. A fair and blind benchmark has likewise been missing. Here we introduce EPIC, the Eukaryotic Promoter and transcription Initiation prediction Challenge. Teams receive strand-specific, single-nucleotide-resolution initiation profiles for 80-95% of the genome and are asked to predict the remainder from DNA sequence alone. To level the field, the challenge relies on understudied animals spanning three phyla: octopus, oyster, milkweed bug, Indian meal moth, and shark. EPIC is open to everyone and closes on December 31, 2026. All teams clearing the dinucleotide precision baseline are invited to join the consortium authorship of the post-challenge publication, and winners are invited for personal authorship.

bioinformatics↗

bioRxiv2026

Classic machine learning on top of multiple position weight matrices improves genomic prediction of transcription factor binding sites

MotivationDNA motifs recognised by transcription factors are typically represented as position weight matrices (PWMs), assuming independent contributions of individual nucleotides to protein binding specificity. Many alternative models accounting for correlations of positional contributions have been introduced in the past decades. However, performance gains have generally not outweighed the advantages of simplicity, interpretability, and practical applicability of PWMs with the well-established codebase. Existing software tools and motif databases provide multiple non-identical PWMs for the same transcription factor or even for the same dataset. It remains a practical question whether these PWMs can be effectively combined into a single improved model. ResultsHere we describe ArChIPelago (https://github.com/autosome-ru/ArChIPelago), a computational framework that combines multiple PWMs into a joint model using classic machine learning techniques, from linear regression to ensembles of decision trees. We show that such a combination improves prediction of transcription factor binding sites in genomic sequences. With a diverse collection of 704 ChIP-Seq datasets spanning 36 orthologous human and mouse transcription factors of diverse structural families, we show that ArChIPelago consistently outperforms the best available individual mono- and dinucleotide PWMs as well as sparse local inhomogeneous mixture models. Furthermore, using both human and mouse data, we demonstrate that PWM ensembles are capable of making reliable cross-species predictions.

bioinformatics↗

bioRxiv2026

A Joint Promoterome-Proteome Atlas Highlights the Molecular Diversity of Human Skeletal Muscles

More than 600 distinct skeletal muscles constitute up to 40% of the total mass of the human body. Human skeletal muscles differ in anatomical position, morphology, origin, and function, but the diversity of their molecular phenotypes, the gene expression and protein abundance profiles, remains poorly explored. Here, we report the large-scale CAGE-Seq promoterome profiling of 75 human skeletal muscles, complemented by 22 matched proteomes obtained with mass spectrometry. We identified 37001 transcribed regulatory elements and 1804 protein groups encompassing 1895 proteins, 80% of which demonstrated non-uniform expression across different muscles. The skeletal muscles of the eye, tongue, and diaphragm had the most distinctive molecular phenotypes, while the overall diversity was driven by hundreds of transcription factors with tissue-specific activity. By analyzing the allelic imbalance of CAGE-Seq reads, we discovered 6653 allele-specific single-nucleotide variants often coinciding with muscle-related GWAS SNPs, including muscle volume. Finally, we provide an interactive online atlas of transcriptomic and proteomic molecular phenotypes, facilitating further studies of gene regulation and heritable pathologies of skeletal muscles.

genomics↗

bioRxiv2025

Transcript-wide m6A methylation defines the efficiency of the cap-independent translation initiation

N6-methyladenosine (m6A) role in translation control and, particularly, in cap-independent initiation has attracted major attention. A common approach is to study the impact of m6A depending on its particular location in the transcripts, but the global impact of m6A along mRNAs remains unclear. Here, we combined ribosome profiling under conditions of mTOR inhibition and m6A mapping to identify distinct subsets of mRNAs that differ in their sensitivity to the suppression of cap-dependent initiation. The sensitivity was strongly correlated with the total m6A methylation of the transcripts. Further, we demonstrated that upon decreased mTOR activity, m6A methylation facilitated the enhanced association of mRNAs with components of the eIF4F complex. Thus, the efficiency of cap-independent translation initiation is primarily defined not by precise localization but by the total level of m6A methylation, and m6A has a compensatory role in maintaining translation when the canonical cap-dependent pathway is impaired. All in all, our findings underscore the significance of contemplating global m6A methylation status as a pivotal element in translational control, particularly under stress or signaling perturbations. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=79 SRC="FIGDIR/small/690160v2_ufig1.gif" ALT="Figure 1"> View larger version (17K): org.highwire.dtl.DTLVardef@e18178org.highwire.dtl.DTLVardef@fb0346org.highwire.dtl.DTLVardef@19406eaorg.highwire.dtl.DTLVardef@1520821_HPS_FORMAT_FIGEXP M_FIG C_FIG

molecular biology↗

bioRxiv2025

Inferring binding specificities of human transcription factors with the wisdom of crowds

DNA motif discovery and, particularly, computational modeling of transcription factor binding motifs, has been a mecca of algorithmic bioinformatics for several decades. Here, we report the results of the largest open community challenge in Inferring BInding Specificities (IBIS), where participants all over the world were invited to construct binding specificity models from multi-assay experimental data for poorly studied human transcription factors. The submissions were rigorously tested against a rich held-out dataset. Benchmarking demonstrated a consistent advantage of properly designed deep learning models over traditional positional weight matrices and other machine learning methods. Yet, the positional weight matrices displayed a surprisingly strong performance out of the box, being only slightly behind the best deep learning models. A post-challenge assessment of a selection of other deep learning methods further solidified this finding. IBIS highlights the power of benchmarking in finding adequate DNA motif representations, emphasizes the pros and cons of various machine learning methods applied to DNA motif modeling, and establishes a rich dataset, benchmarking protocols, and computational framework for a fair cross-platform evaluation of future models of transcription factor binding motifs in DNA sequences. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=175 SRC="FIGDIR/small/688692v1_ufig1.gif" ALT="Figure 1"> View larger version (64K): org.highwire.dtl.DTLVardef@1c6677corg.highwire.dtl.DTLVardef@b4124aorg.highwire.dtl.DTLVardef@1ce2b1org.highwire.dtl.DTLVardef@66e917_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2024

A generative framework for enhanced cell-type specificity in rationally designed mRNAs

mRNA delivery offers new opportunities for disease treatment by directing cells to produce therapeutic proteins. However, designing highly stable mRNAs with programmable cell type-specificity remains a challenge. To address this, we measured the regulatory activity of 60,000 5 and 3 untranslated regions (UTRs) across six cell types and developed PARADE (Prediction And RAtional DEsign of mRNA UTRs), a generative AI framework to engineer untranslated RNA regions with tailored cell type-specific activity. We validated PARADE by testing 15,800 de novo-designed sequences across these cell lines and identified many sequences that demonstrated superior specificity and activity compared to existing RNA therapeutics. mRNAs with PARADE-engineered UTRs also exhibited robust tissue-specific activity in animal models, achieving selective expression in the liver and spleen. We also leveraged PARADE to enhance mRNA stability, significantly increasing protein output and therapeutic durability in vivo. These advancements translated to notable increases in therapeutic efficacy, as PARADE-designed UTRs in oncosuppressor mRNAs, namely PTEN and P16, effectively reduced tumor growth in patient-derived neuroglioma xenograft models and orthotopic mouse models. Collectively, these findings establish PARADE as a versatile platform for designing safer, more precise, and highly stable mRNA therapies.

systems biology↗

bioRxiv2024

GHT-SELEX demonstrates unexpectedly high intrinsic sequence specificity and complex DNA binding of many human transcription factors

Precise identification of transcription factor (TF) binding sites is a long standing challenge in human regulatory genomics: TF binding motifs are short and degenerate, while the genome is large. Motif scans, therefore, often produce excessive binding site predictions. By surveying 179 TFs across 25 families using >1,500 cyclic in vitro selection experiments with fragmented, naked, and unmodified genomic DNA - a method we term GHT-SELEX (Genomic HT-SELEX) - we find that many human TFs possess much higher sequence specificity than anticipated. Moreover, genomic binding regions from GHT-SELEX are often surprisingly similar to those obtained in vivo (i.e., ChIP-seq peaks). Contrary to conventional wisdom, we find that high specificity can also be obtained from motif scans, but performance is highly dependent on the derivation and use of the motifs, including accounting for multiple local matches. We also observe alternative engagement of multiple DNA-binding domains within the same protein: long C2H2 zinc finger proteins often utilize modular DNA recognition, engaging different subsets of their DNA-binding domain (DBD) arrays to recognize multiple types of distinct target sites, frequently evolving via internal duplication and divergence of one or more DBDs. Thus, it is common for TFs to possess sufficient intrinsic specificity to delineate a large fraction of in vivo genomic targets, independently of other cellular factors.

genomics↗

bioRxiv2024

Cross-platform DNA motif discovery and benchmarking to explore binding specificities of poorly studied human transcription factors

A DNA sequence pattern, or "motif", is an essential representation of DNA-binding specificity of a transcription factor (TF). Any particular motif model has potential flaws due to shortcomings of the underlying experimental data and computational motif discovery algorithm. As a part of the Codebook/GRECO-BIT initiative, here we evaluated at large scale the cross-platform recognition performance of positional weight matrices (PWMs), which remain popular motif models in many practical applications. We applied ten different DNA motif discovery tools to generate PWMs from the "Codebook" data comprised of 4,237 experiments from five different platforms profiling the DNA-binding specificity of 394 human proteins, focusing on understudied transcription factors of different structural families. For many of the proteins, there was no prior knowledge of a genuine motif. By benchmarking-supported human curation, we constructed an approved subset of experiments comprising about 30% of all experiments and 50% of tested TFs which displayed consistent motifs across platforms and replicates. We present the Codebook Motif Explorer (https://mex.autosome.org), a detailed online catalog of DNA motifs, including the top-ranked PWMs, and the underlying source and benchmarking data. We demonstrate that in the case of high-quality experimental data, most of the popular motif discovery tools detect valid motifs and generate PWMs, which perform well both on genomic and synthetic data. Yet, for each of the algorithms, there were problematic combinations of proteins and platforms, and the basic motif properties such as nucleotide composition and information content offered little help in detecting such pitfalls. By combining multiple PMWs in decision trees, we demonstrate how our setup can be readily adapted to train and test binding specificity models more complex than PWMs. Overall, our study provides a rich motif catalog as a solid baseline for advanced models and highlights the power of the multi-platform multi-tool approach for reliable mapping of DNA binding specificities. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=141 SRC="FIGDIR/small/619379v2_ufig1.gif" ALT="Figure 1"> View larger version (61K): org.highwire.dtl.DTLVardef@79561forg.highwire.dtl.DTLVardef@54c0aorg.highwire.dtl.DTLVardef@1c33f34org.highwire.dtl.DTLVardef@16a93ba_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical AbstractC_FLOATNO C_FIG

bioinformatics↗

bioRxiv2024

Identification of methylation-sensitive human transcription factors using meSMiLE-seq

Transcription factors (TFs) are key players in eukaryotic gene regulation, but the DNA binding specificity of many TFs remains unknown. Here, we assayed 284 mostly poorly characterized, putative human TFs using selective microfluidics-based ligand enrichment followed by sequencing (SMiLE-seq), revealing 72 new DNA binding motifs. To investigate whether some of the 158 TFs for which we did not find motifs preferably bind epigenetically modified DNA (i.e. methylated CG dinucleotides), we developed methylation-sensitive SMiLE-seq (meSMiLE-seq). This microfluidic assay simultaneously probes the affinity of a protein to methylated and unmethylated DNA, augmenting the capabilities of the original method to infer methylation-aware binding sites. We assayed 114 TFs with meSMiLE-seq and identified DNA-binding models for 48 proteins, including the known methylation-sensitive binding modes for POU5F1 and RFX5. For 11 TFs, binding to methylated DNA was preferred or resulted in the discovery of alternative, methylation-dependent motifs (e.g. PRDM13), while aversion towards methylated sequences was found for 13 TFs (e.g. USF3). Finally, we uncovered a potential role for ZHX2 as a putative binder of Z-DNA, a left-handed helical DNA structure which is adopted more frequently upon CpG methylation. Altogether, our study significantly expands the human TF codebook by identifying DNA binding motifs for 98 TFs, while providing a versatile platform to quantitatively assay the impact of DNA modifications on TF binding.

genomics↗

bioRxiv2024

Perspectives on Codebook: sequence specificity of uncharacterized human transcription factors

Gene expression is regulated by transcription factors (TFs), which recognize specific DNA sequence motifs. Several hundred putative human TFs, identified mainly by an apparent DNA-binding domain, lack known binding motifs1, and even for well-characterized TFs, it remains controversial to what degree motifs accurately reflect binding sites in living cells2,3. Here, we describe a systematic effort ("Codebook") to determine the sequence specificity of 332 putative and poorly characterized human TFs. Over 4,000 independent experiments, encompassing multiple in vitro and in vivo assays, produced motifs for just over half (177, or 53%), of which most are unique to a single protein, thereby extending the vocabulary of sequence recognition encoded by human TFs by [~]100 distinct motifs. Moreover, binding motifs identified in vitro are strongly enriched within cellular binding sites. Collectively, the data reveal tens of thousands of previously unknown, conserved, and direct TF binding sites across the human genome. These sites are concentrated in promoter regions, and are predictive of gene expression, illustrating that this new data atlas provides an important step forward in decoding the human genome.

genomics↗

bioRxiv2024

Extensive binding of uncharacterized human transcription factors to genomic dark matter

The functional impact of a large portion of the human genome known as "dark matter DNA", which is composed mainly of repeat sequences, remains enigmatic. The genome also encodes hundreds of putative and poorly characterized transcription factors (TFs). Here, we determined genomic binding locations of 166 poorly characterized human TFs in living cells. Nearly half of them associate strongly with known regulatory regions such as promoters and enhancers, frequently co-localizing with each other at conserved motif matches. The other half often associate with genomic dark matter, however, at largely non-overlapping (i.e., unique) sites, via intrinsic sequence recognition. Fifty-four of the latter half, which we term "Dark TFs", mainly bind within regions of closed chromatin, with each recognizing a unique set of repeat sequences. The Dark TFs include many KZNFs, which are known to bind and silence TEs, and other TFs with apparent repressive functions. By contrast, some may be pioneers: we find that induction of TPRX1, a known regulator of zygotic preimplantation, leads to chromatin opening at many of its binding sites in the dark matter genome. Altogether, our results shed light on a large fraction of poorly characterized human TFs and simultaneously illuminate the diversity of function within the dark matter genome.

genomics↗

bioRxiv2023

Statistical framework for calling allelic imbalance in high-throughput sequencing data

High-throughput sequencing facilitates large-scale studies of gene regulation and allows tracing the associations of individual genomic variants with changes in gene expression. Compared to classic association studies, allelic imbalance at heterozygous variants captures the functional effects of the regulatory genome variation with smaller sample sizes and higher sensitivity. Yet, the identification of allele-specific events from allelic read counts remains non-trivial due to multiple sources of technical and biological variability, which induce data-dependent biases and overdispersion. Here we present MIXALIME, a novel computational framework for calling allele-specific events in diverse omics data with a repertoire of statistical models accounting for read mapping bias and copy-number variation. We benchmark MIXALIME against existing tools and demonstrate its practical usage by constructing an atlas of allele-specific chromatin accessibility, UDACHA, from thousands of available datasets obtained from diverse cell types. Availabilityhttps://github.com/autosome-ru/MixALime, https://udacha.autosome.org

bioinformatics↗

bioRxiv2023

Evaluation and optimization of sequence-based gene regulatory deep learning models

Neural networks have emerged as immensely powerful tools in predicting functional genomic regions, notably evidenced by recent successes in deciphering gene regulatory logic. However, a systematic evaluation of how model architectures and training strategies impact genomics model performance is lacking. To address this gap, we held a DREAM Challenge where competitors trained models on a dataset of millions of random promoter DNA sequences and corresponding expression levels, experimentally determined in yeast, to best capture the relationship between regulatory DNA and gene expression. For a robust evaluation of the models, we designed a comprehensive suite of benchmarks encompassing various sequence types. While some benchmarks produced similar results across the top-performing models, others differed substantially. All top-performing models used neural networks, but diverged in architectures and novel training strategies, tailored to genomics sequence data. To dissect how architectural and training choices impact performance, we developed the Prix Fixe framework to divide any given model into logically equivalent building blocks. We tested all possible combinations for the top three models and observed performance improvements for each. The DREAM Challenge models not only achieved state-of-the-art results on our comprehensive yeast dataset but also consistently surpassed existing benchmarks on Drosophila and human genomic datasets. Overall, we demonstrate that high-quality gold-standard genomics datasets can drive significant progress in model development.

genomics↗

bioRxiv2023

Systematic Identification of Post-Transcriptional Regulatory Modules

In our cells, a limited number of RNA binding proteins (RBPs) are responsible for all aspects of RNA metabolism across the entire transcriptome. To accomplish this, RBPs form regulatory units that act on specific target regulons. However, the landscape of RBP combinatorial interactions remains poorly explored. Here, we performed a systematic annotation of RBP combinatorial interactions via multimodal data integration. We built a large-scale map of RBP protein neighborhoods by generating in vivo proximity-dependent biotinylation datasets of 50 human RBPs. In parallel, we used CRISPR interference with single-cell readout to capture transcriptomic changes upon RBP knockdowns. By combining these physical and functional interaction readouts, along with the atlas of RBP mRNA targets from eCLIP assays, we generated an integrated map of functional RBP interactions. We then used this map to match RBPs to their context-specific functions and validated the predicted functions biochemically for four RBPs. This study highlights the previously underappreciated scale of the inter-RBP interactions, be it genetic or physical, and is a first step towards a more comprehensive understanding of post-transcriptional regulatory processes and their underlying molecular grammar.

systems biology↗

bioRxiv2023

Human tissues exhibit diverse composition of translation machinery

While protein synthesis is vital for the majority of cell types of the human body, diversely differentiated cells require specific translation regulation. This suggests specialization of translation machinery across tissues and organs. Using transcriptomic data from GTEx, FANTOM, and Gene Atlas we systematically explored the abundance of transcripts encoding translation factors and aminoacyl-tRNA synthetases (ARSases) in human tissues. We revised a few known and identified several novel translation-related genes exhibiting strict tissue-specific expression. The proteins they encode include eEF1A1, eEF1A2, PABPC1L, PABPC3, eIF1B, eIF4E1B, eIF4ENIF1, and eIF5AL1. Furthermore, our analysis revealed a pervasive tissue-specific relative abundance of translation machinery components (e.g. PABP and eRF3 paralogs, eIF2B subunits, eIF5MPs, and some ARSases), suggesting presumptive variance in the composition of translation initiation, elongation, and termination complexes. These conclusions were largely confirmed by the analysis of proteomic data. Finally, we paid attention to sexual dimorphism in the repertoire of translation factors encoded in sex chromosomes (eIF1A, eIF2{gamma}, and DDX3), and identified testis and brain as organs with the most diverged expression of translation-associated genes.

molecular biology↗

bioRxiv2022

LegNet: resetting the bar in deep learning for accurate prediction of promoter activity and variant effects from massive parallel reporter assays

MotivationThe increasing volume of data from high-throughput experiments including parallel reporter assays facilitates the development of complex deep learning approaches for DNA regulatory grammar. ResultsHere we introduce LegNet, an EfficientNetV2-inspired convolutional network for modeling short gene regulatory regions. By approaching the sequence-to-expression regression problem as a soft classification task, LegNet secured first place for the autosome.org team in the DREAM 2022 challenge of predicting gene expression from gigantic parallel reporter assays. Using published data, here we demonstrate that LegNet outperforms existing models and accurately predicts gene expression per se as well as the effects of single-nucleotide variants. Furthermore, we show how LegNet can be used in a diffusion network manner for the rational design of promoter sequences yielding the desired expression level. Availability and Implementationhttps://github.com/autosome-ru/LegNet. The GitHub repository includes the Python code under the MIT license to reproduce the results presented in the study and a Jupyter Notebook tutorial. Supplementary InformationOnline-only supplementary data are available at Bioinformatics online. Contactdmitrypenzar1996@gmail.com, ivan.kulakovskiy@gmail.com

bioinformatics↗

bioRxiv2021

A standard knockout procedure alters expression of adjacent loci at the translational level

The S. cerevisiae gene deletion collection is widely used for functional gene annotation and genetic interaction analyses. However, the standard G418-resistance cassette used to produce knockout mutants delivers strong regulatory elements into the target genetic loci. To date, its side effects on the expression of neighboring genes have never been systematically assessed. Here, using ribosome profiling data, RT-qPCR, and reporter expression, we investigated perturbations induced by the KanMX module. Our analysis revealed significant alterations in the transcription efficiency of neighboring genes and, more importantly, severe impairment of their mRNA translation, leading to changes in protein abundance. In the "head-to-head" orientation of the neighbor and the deleted gene, knockout often led to a shift of the transcription start site of the neighboring gene, introducing new uAUG codon(s) into the expanded 5 untranslated region (5 UTR). In the "tail-to-tail" arrangement, knockout led to activation of alternative polyadenylation signals in the neighboring gene, thus altering its 3 UTR. These events may explain the so-called neighboring gene effect (NGE), i.e. false genetic interactions of the deleted genes. We estimate that in as much as [~]1/5 of knockout strains the expression of neighboring genes may be substantially (>2-fold) deregulated at the level of translation.

molecular biology↗

bioRxiv2021

Functional annotation of human long noncoding RNAs using chromatin conformation data

The human genome is pervasively transcribed and produces a wide variety of long non-coding RNAs (lncRNAs), constituting the majority of transcripts across human cell types. Studying lncRNAs is challenging due to their low expression level, cell type-specific occurrence, poor sequence conservation between orthologs, and lack of information about RNA domains. LncRNAs direct the regulatory factors in the locations that are in cis to their transcription sites. We designed a model to predict if an lncRNA acts in cis based on its features and trained it using RNA-chromatin interaction data. The trained model is cell type-independent and does not require RNA-chromatin data. Combining RNA-chromatin and Hi-C data, we showed that lncRNA-chromatin binding sites are determined by chromosome conformation. For each lncRNA, the spatially proximal genes were identified as their potential targets by combining Hi-C and Cap Analysis Gene Expression (CAGE) data in 18 human cell types. RNA-protein and RNA-chromatin interaction data suggested that lncRNAs act as scaffolds to recruit regulatory proteins to target promoters and enhancers. We provide the data through an interactive visualization web portal at https://fantom.gsc.riken.jp/zenbu/reports/#F6_3D_lncRNA.

genomics↗