Search

Single-nucleotide variants in enhancers or promoters may affect gene transcription by altering transcription factor binding sites. Here the authors present a meta-analysis empowered by a new statistical method covering thousands of ChIP-Seq experiments resulting in the identification of more than 500 thousand allele-specific binding (ASB) events in the human genome.

Many problems of modern genetics and functional genomics require the assessment of functional effects of sequence variants, including gene expression changes. Machine learning is considered to be a promising approach for solving this task, but its practical applications remain a challenge due to the insufficient volume and diversity of training data. A promising source of valuable data is a saturation mutagenesis massively parallel reporter assay, which quantitatively measures changes in transcription activity caused by sequence variants. Here, we explore the computational predictions of the effects of individual single-nucleotide variants on gene transcription measured in the massively parallel reporter assays, based on the data from the recent “Regulation Saturation” Critical Assessment of Genome Interpretation challenge. We show that the estimated prediction quality strongly depends on the structure of the training and validation data. Particularly, training on the sequence segments located next to the validation data results in the “information leakage” caused by the local context. This information leakage allows reproducing the prediction quality of the best CAGI challenge submissions with a fairly simple machine learning approach, and even obtaining notably better-than-random predictions using irrelevant genomic regions. Validation scenarios preventing such information leakage dramatically reduce the measured prediction quality. The performance at independent regulatory regions entirely excluded from the training set appears to be much lower than needed for practical applications, and even the performance estimation will become reliable only in the future with richer data from multiple reporters. The source code and data are available at https://bitbucket.org/autosomeru_cagi2018/cagi2018_regsat and https://genomeinterpretation.org/content/expression-variants.

The human genome contains millions of candidate cis-regulatory elements (cCREs) with cell-type-specific activities that shape both health and many disease states 1 . However, we lack a functional understanding of the sequence features that control the activity and cell-type-specific features of these cCREs. Here we used lentivirus-based massively parallel reporter assays (lentiMPRAs) to test the regulatory activity of more than 680,000 sequences, representing an extensive set of annotated cCREs among three cell types (HepG2, K562 and WTC11), and found that 41.7% of these sequences were active. By testing sequences in both orientations, we find promoters to have strand-orientation biases and their 200-nucleotide cores to function as non-cell-type-specific 'on switches' that provide similar expression levels to their associated gene. By contrast, enhancers have weaker orientation biases, but increased tissue-specific characteristics. Utilizing our lentiMPRA data, we develop sequence-based models to predict cCRE function and variant effects with high accuracy, delineate regulatory motifs and model their combinatorial effects. Testing a lentiMPRA library encompassing 60,000 cCREs in all three cell types further identified factors that determine cell-type specificity. Collectively, our work provides an extensive catalogue of functional CREs in three widely used cell lines and showcases how large-scale functional measurements can be used to dissect regulatory grammar., Competing Interests: Competing interests: V.A. is an employee of Sanofi, but performed this work independently of Sanofi. J.S. is a scientific advisory board member, consultant and/or co-founder of Cajal Neuroscience, Guardant Health, Maze Therapeutics, Camp4 Therapeutics, Phase Genomics, Adaptive Biotechnologies, Scale Biosciences, Somite Therapeutics, Sixth Street Capital and Pacific Biosciences. N.A. is a co-founder and on the scientific advisory board of Regel Therapeutics and receives funding from BioMarin Pharmaceutical Incorporated. All other authors declare no competing interests., (© 2025. The Author(s).)

Recent advancements in genomics, propelled by artificial intelligence, have unlocked unprecedented capabilities in interpreting genomic sequences, mitigating the need for exhaustive experimental analysis of complex, intertwined molecular processes inherent in DNA function. A significant challenge, however, resides in accurately decoding genomic sequences, which inherently involves comprehending rich contextual information dispersed across thousands of nucleotides. To address this need, we introduce GENA language model (GENA-LM), a suite of transformer-based foundational DNA language models capable of handling input lengths up to 36 000 base pairs. Notably, integrating the newly developed recurrent memory mechanism allows these models to process even larger DNA segments. We provide pre-trained versions of GENA-LM, including multispecies and taxon-specific models, demonstrating their capability for fine-tuning and addressing a spectrum of complex biological tasks with modest computational demands. While language models have already achieved significant breakthroughs in protein biology, GENA-LM showcases a similarly promising potential for reshaping the landscape of genomics and multi-omics data analysis. All models are publicly available on GitHub (https://github.com/AIRI-Institute/GENA_LM) and on HuggingFace (https://huggingface.co/AIRI-Institute). In addition, we provide a web service (https://dnalm.airi.net/) allowing user-friendly DNA annotation with GENA-LM models., (© The Author(s) 2025. Published by Oxford University Press on behalf of Nucleic Acids Research.)

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., Competing Interests: Competing Interests M.K. and H.G. are inventors on a provisional patent related to this study. The remaining authors declare no competing interests.

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., Competing Interests: Competing interests O.F. is employed by Roche.

A systematic evaluation of how model architectures and training strategies impact genomics model performance is needed. 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. For a robust evaluation of the models, we designed a comprehensive suite of benchmarks encompassing various sequence types. All top-performing models used neural networks but diverged in architectures and training strategies. To dissect how architectural and training choices impact performance, we developed the Prix Fixe framework to divide models into modular building blocks. We tested all possible combinations for the top three models, further improving their performance. 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, demonstrating the progress that can be driven by gold-standard genomics datasets., (© 2024. The Author(s).)

Prediction of RNA structure from sequence remains an unsolved problem, and progress has been slowed by a paucity of experimental data. Here, we present Ribonanza, a dataset of chemical mapping measurements on two million diverse RNA sequences collected through Eterna and other crowdsourced initiatives. Ribonanza measurements enabled solicitation, training, and prospective evaluation of diverse deep neural networks through a Kaggle challenge, followed by distillation into a single, self-contained model called RibonanzaNet. When fine tuned on auxiliary datasets, RibonanzaNet achieves state-of-the-art performance in modeling experimental sequence dropout, RNA hydrolytic degradation, and RNA secondary structure, with implications for modeling RNA tertiary structure.

Phylogenetic inference based on protein sequence alignment is a widely used procedure. Numerous phylogenetic algorithms have been developed, most of which have many parameters and options. Choosing a program, options, and parameters can be a nontrivial task. No benchmark for comparison of phylogenetic programs on real protein sequences was publicly available. We have developed PhyloBench, a benchmark for evaluating the quality of phylogenetic inference, and used it to test a number of popular phylogenetic programs. PhyloBench is based on natural, not simulated, protein sequences of orthologous evolutionary domains. The measure of accuracy of an inferred tree is its distance to the corresponding species tree. A number of tree-to-tree distance measures were tested. The most reliable results were obtained using the Robinson-Foulds distance. Our results confirmed recent findings that distance methods are more accurate than maximum likelihood (ML) and maximum parsimony. We tested the bayesian program MrBayes on natural protein sequences and found that, on our datasets, it performs better than ML, but worse than distance methods. Of the methods we tested, the Balanced Minimum Evolution method implemented in FastME yielded the best results on our material. Alignments and reference species trees are available at https://mouse.belozersky.msu.ru/tools/phylobench/ together with a web-interface that allows for a semi-automatic comparison of a user's method with a number of popular programs., Competing Interests: Conflict of interest statement. None declared., (© The Author(s) 2024. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution.)

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., Competing Interests: Competing interests E.D.V is the founder of Sequome, Inc. A.R. is an employee of Genentech and has equity in Roche. A.R. is a co-founder and equity holder of Celsius Therapeutics, an equity holder in Immunitas and, until 31 July 2020, was a scientific advisory board member of Thermo Fisher Scientific, Syros Pharmaceuticals, Neogene Therapeutics and Asimov. A.R. was an Investigator of the Howard Hughes Medical Institute when this work was initiated. The remaining authors declare no competing interests.

Motivation: The increasing volume of data from high-throughput experiments including parallel reporter assays facilitates the development of complex deep-learning approaches for modeling DNA regulatory grammar., Results: Here, 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 Implementation: https://github.com/autosome-ru/LegNet. The GitHub repository includes Jupyter Notebook tutorials and Python scripts under the MIT license to reproduce the results presented in the study., (© The Author(s) 2023. Published by Oxford University Press.)

Background: Positional weight matrix (PWM) is a de facto standard model to describe transcription factor (TF) DNA binding specificities. PWMs inferred from in vivo or in vitro data are stored in many databases and used in a plethora of biological applications. This calls for comprehensive benchmarking of public PWM models with large experimental reference sets., Results: Here we report results from all-against-all benchmarking of PWM models for DNA binding sites of human TFs on a large compilation of in vitro (HT-SELEX, PBM) and in vivo (ChIP-seq) binding data. We observe that the best performing PWM for a given TF often belongs to another TF, usually from the same family. Occasionally, binding specificity is correlated with the structural class of the DNA binding domain, indicated by good cross-family performance measures. Benchmarking-based selection of family-representative motifs is more effective than motif clustering-based approaches. Overall, there is good agreement between in vitro and in vivo performance measures. However, for some in vivo experiments, the best performing PWM is assigned to an unrelated TF, indicating a binding mode involving protein-protein cooperativity., Conclusions: In an all-against-all setting, we compute more than 18 million performance measure values for different PWM-experiment combinations and offer these results as a public resource to the research community. The benchmarking protocols are provided via a web interface and as docker images. The methods and results from this study may help others make better use of public TF specificity models, as well as public TF binding data sets.

Background: Transposons are selfish genetic elements that self-reproduce in host DNA. They were active during evolutionary history and now occupy almost half of mammalian genomes. Close insertions of transposons reshaped structure and regulation of many genes considerably. Co-evolution of transposons and host DNA frequently results in the formation of new regulatory regions. Previously we published a concept that the proportion of functional features held by transposons positively correlates with the rate of regulatory evolution of the respective genes., Methods: We ranked human genes and molecular pathways according to their regulatory evolution rates based on high throughput genome-wide data on five histone modifications (H3K4me3, H3K9ac, H3K27ac, H3K27me3, H3K9me3) linked with transposons for five human cell lines., Results: Based on the total of approximately 1.5 million histone tags, we ranked regulatory evolution rates for 25075 human genes and 3121 molecular pathways and identified groups of molecular processes that showed signs of either fast or slow regulatory evolution. However, histone tags showed different regulatory patterns and formed two distinct clusters: promoter/active chromatin tags (H3K4me3, H3K9ac, H3K27ac) vs. heterochromatin tags (H3K27me3, H3K9me3)., Conclusion: In humans, transposon-linked histone marks evolved in a coordinated way depending on their functional roles.

Background: Retroelements (REs) are transposable elements occupying ~40% of the human genome that can regulate genes by providing transcription factor binding sites (TFBS). RE-linked TFBS profile can serve as a marker of gene transcriptional regulation evolution. This approach allows for interrogating the regulatory evolution of organisms with RE-rich genomes. We aimed to characterize the evolution of transcriptional regulation for human genes and molecular pathways using RE-linked TFBS accumulation as a metric. Methods: We characterized human genes and molecular pathways either enriched or deficient in RE-linked TFBS regulation. We used ENCODE database with mapped TFBS for 563 transcription factors in 13 human cell lines. For 24,389 genes and 3124 molecular pathways, we calculated the score of RE-linked TFBS regulation reflecting the regulatory evolution rate at the level of individual genes and molecular pathways. Results: The major groups enriched by RE regulation deal with gene regulation by microRNAs, olfaction, color vision, fertilization, cellular immune response, and amino acids and fatty acids metabolism and detoxication. The deficient groups were involved in translation, RNA transcription and processing, chromatin organization, and molecular signaling. Conclusion: We identified genes and molecular processes that have characteristics of especially high or low evolutionary rates at the level of RE-linked TFBS regulation in human lineage.

Background: Many algorithms and programs are available for phylogenetic reconstruction of families of proteins. Methods used widely at present use either a number of distance-based principles or character-based principles of maximum parsimony or maximum likelihood., Results: We developed a novel program, named PQ, for reconstructing protein and nucleic acid phylogenies following a new character-based principle. Being tested on natural sequences PQ improves upon the results of maximum parsimony and maximum likelihood. Working with alignments of 10 and 15 sequences, it also outperforms the FastME program, which is based on one of the distance-based principles. Among all tested programs PQ is proved to be the least susceptible to long branch attraction. FastME outperforms PQ when processing alignments of 45 sequences, however. We confirm a recent result that on natural sequences FastME outperforms maximum parsimony and maximum likelihood. At the same time, both PQ and FastME are inferior to maximum parsimony and maximum likelihood on simulated sequences. PQ is open source and available to the public via an online interface., Conclusions: The software we developed offers an open-source alternative for phylogenetic reconstruction for relatively small sets of proteins and nucleic acids, with up to a few tens of sequences.

Endogenous retroviruses and retrotransposons also termed retroelements (REs) are mobile genetic elements that were active until recently in human genome evolution. REs regulate gene expression by actively reshaping chromatin structure or by directly providing transcription factor binding sites (TFBSs). We aimed to identify molecular processes most deeply impacted by the REs in human cells at the level of TFBS regulation. By using ENCODE data, we identified ~2 million TFBS overlapping with putatively regulation-competent human REs located in 5-kb gene promoter neighborhood (~17% of all TFBS in promoter neighborhoods; ~9% of all RE-linked TFBS). Most of REs hosting TFBS were highly diverged repeats, and for the evolutionary young (0-8% diverged) elements we identified only ~7% of all RE-linked TFBS. The gene-specific distributions of RE-linked TFBS generally correlated with the distributions for all TFBS. However, several groups of molecular processes were highly enriched in the RE-linked TFBS regulation. They were strongly connected with the immunity and response to pathogens, with the negative regulation of gene transcription, ubiquitination, and protein degradation, extracellular matrix organization, regulation of STAT signaling, fatty acids metabolism, regulation of GTPase activity, protein targeting to Golgi, regulation of cell division and differentiation, development and functioning of perception organs and reproductive system. By contrast, the processes most weakly affected by the REs were linked with the conservative aspects of embryo development. We also identified differences in the regulation features by the younger and older fractions of the REs. The regulation by the older fraction of the REs was linked mainly with the immunity, cell adhesion, cAMP, IGF1R, Notch, Wnt, and integrin signaling, neuronal development, chondroitin sulfate and heparin metabolism, and endocytosis. The younger REs regulate other aspects of immunity, cell cycle progression and apoptosis, PDGF, TGF beta, EGFR, and p38 signaling, transcriptional repression, structure of nuclear lumen, catabolism of phospholipids, and heterocyclic molecules, insulin and AMPK signaling, retrograde Golgi-ER transport, and estrogen signaling. The immunity-linked pathways were highly represented in both categories, but their functional roles were different and did not overlap. Our results point to the most quickly evolving molecular pathways in the recent and ancient evolution of human genome.