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.

Background: The three-dimensional organization of the genome is tightly connected to its biological function. The Hi-C approach was recently introduced as a method that can be used to identify higher-order chromatin interactions genome-wide. The aim of this study was to determine genome-wide chromatin interaction frequencies using the Hi-C approach in mouse sperm cells and embryonic fibroblasts. Results: The obtained data demonstrate that the three-dimensional genome organizations of sperm and fibroblast cells show a high degree of similarity both with each other and with the previously described mouse embryonic stem cells. Both A- and B-compartments and topologically associated domains are present in spermatozoa and fibroblasts. Nevertheless, sperm cells and fibroblasts exhibit statistically significant differences between each other in the contact probabilities of defined loci. Tight packaging of the sperm genome results in an enrichment of long-range contacts compared with the fibroblasts. However, only 30% of the differences in the number of contacts are based on differences in the densities of their genome packages; the main source of the differences is the gain or loss of contacts that are specific for defined genome regions. We find that the dependence of the contact probability on genomic distance for sperm is close to the dependence predicted for the fractal globular folding of chromatin. Conclusions: Overall, we can conclude that the three-dimensional structure of the genome is passed through generations without being dramatically changed in sperm cells.

How chromosomes are folded, spatially organized and regulated in three dimensions inside the cell nucleus are among the longest standing questions in cell biology. Genome-wide chromosome conformation capture (Hi-C) technique allowed identifying and characterizing spatial chromatin compartments in several mammalian species. Here, we present the first genome-wide analysis of chromatin interactions in chicken embryonic fibroblasts (CEF) and adult erythrocytes. We showed that genome of CEF is partitioned into topologically associated domains (TADs), distributed in accordance with gene density, transcriptional activity and CTCF-binding sites. In contrast to mammals, where all examined somatic cell types display relatively similar spatial organization of genome, chicken erythrocytes strongly differ from fibroblasts, showing pronounced A- and B- compartments, absence of typical TADs and formation of long-range chromatin interactions previously observed on mitotic chromosomes. Comparing mammalian and chicken genome architectures, we provide evidence highlighting evolutionary role of chicken TADs and their significance in genome activity and regulation.

Recent experimental and computational efforts have provided large data sets describing three-dimensional organization of mouse and human genomes and showed the interconnection between the expression profile, epigenetic state, and spatial interactions of loci. These interconnections were utilized to infer the spatial organization of chromatin, including enhancer–promoter contacts, from one-dimensional epigenetic marks. Here, we show that the predictive power of some of these algorithms is overestimated due to peculiar properties of the biological data. We propose an alternative approach, which provides high-quality predictions of chromatin interactions using information on gene expression and CTCF-binding alone. Using multiple metrics, we confirmed that our algorithm could efficiently predict the three-dimensional architecture of both normal and rearranged genomes.

Chromosomes are hierarchically folded within cell nuclei into territories, domains and subdomains, but the functional importance and evolutionary dynamics of these hierarchies are poorly defined. Here, we comprehensively profile genome organizations of five Anopheles mosquito species and show how different levels of chromatin architecture influence each other. Patterns observed on Hi-C maps are associated with known cytological structures, epigenetic profiles, and gene expression levels. Evolutionary analysis reveals conservation of chromatin architecture within synteny blocks for tens of millions of years and enrichment of synteny breakpoints in regions with increased genomic insulation. However, in-depth analysis shows a confounding effect of gene density on both insulation and distribution of synteny breakpoints, suggesting limited causal relationship between breakpoints and regions with increased genomic insulation. At the level of individual loci, we identify specific, extremely long-ranged looping interactions, conserved for ~100 million years. We demonstrate that the mechanisms underlying these looping contacts differ from previously described Polycomb-dependent interactions and clustering of active chromatin.

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-LM, a suite of transformer-based foundational DNA language models capable of handling input lengths up to 36,000 base pairs. Notably, integration of the newly-developed Recurrent Memory mechanism allows these models to process even larger DNA segments. We provide pre-trained versions of GENA-LM, 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 HuggingFace https://huggingface.co/AIRI-Institute.