Biology subjects

Poptsova, M.

Publications and source records attributed to Poptsova, M..

12 recordsLinked to original sources

bioRxiv2025

Predicting RNA:DNA Triplex Structures from Sequence Features Using Deep Learning Architectures

Long non-coding RNAs (lncRNAs) can perform their regulatory roles by forming triple helices through RNA-DNA interactions. Although this has been verified by a few in vivo and in vitro methods, robust in silico approaches that predict the potential of lncRNAs and DNA sites to form triplex structures are still required. Tools such as Triplexator have predicted vast numbers of lncRNAs and DNA sites with triplex forming potential, yet there remains a pressing need for advanced computational methods that can refine and extend these predictions. In this study, we developed ten (10) deep neural network models that predict the potential of lncRNAs and DNA sites to form triple helices on a genome-wide scale. To prepare our dataset, we first used Triplexator to screen out lncRNAs and DNA sites with low triplex-forming potential. We then trained different deep learning architectures, including two-layer convolutional neural networks (CNN), residual neural networks (ResNN), long short-term memory recurrent neural networks (LSTM-RNN), and multilayer perceptron (MLP). Among these architectures, our lncRNA_CNN and LSTM3-RNN both achieved a mean AUC of 0.99 for lncRNA features at a kernel size of 32 and a learning rate of 1e-3. For DNA site features, our DNA_CNN achieved the best performance with a mean AUC of 0.98 under the same conditions. In conclusion, we demonstrate that deep neural network architectures can effectively learn sequence features of lncRNAs and DNA to accurately predict RNA:DNA triplex formation potential, providing a scalable in silico framework for studying genome-wide triplex biology.

bioinformatics↗

bioRxiv2025

Using Deep Learning with Different Architectures to Recognize RNA:DNA Triplex Structures from Histone Modification Features

Long non-coding RNAs (lncRNAs) can perform their regulatory roles by forming triple helices through RNA-DNA interaction. Although this has been verified by few in vivo and in vitro methods, in silico approaches that seek to predict the potentials of lncRNAs and DNA sites becoming a triplex forming structure is required. Triplexator have also predicted vast amounts of lncRNAs and DNA sites that has the potentials of becoming a triplex structure. There is also an emerging experimental-evidence that the presence of epigenetic marks at DNA sites and lncRNAs can facilitate the formation of RNA:DNA triplex structures. There is therefore, a huge demand for computati onal approaches such as deep learning that can make novel predictions about RNA:DNA triplex structure formation. In this study, we developed four (4) deep neural network models that can predict the potentials of lncRNAs and DNA sites to form triple helices genome-wide, by taking histone modification marks as features. Our data was first passed through the Triplexator to screen out lncRNAs and DNA sites with low potentials of forming triple helices. We used different deep learning architectures to build our models, including two-layer convolutional neural networks (CNN) and multilayer perceptron (MLP). Our DNA2_CNN model performed best at a mean AUC of 0.78 at 32 Kernel size and learning rate of 1e-3. Our deep neural network models revealed several novel lncRNAs and DNA sites, including HOTAIR, MEG3, PARTICLE, DACOR1, MIR100HG, FENDRR, ANRIL, TUG1, MALAT1, LINC00599, TINCR, NEAT1, roX2, DHFR, OTX2-AS1, Xist, SNHG16, ATXN8OS, BCYRN1, TERC, Khps1, that have the potential of forming triplex structures, thereby confirming previous experimental results and that of the Triplexator. The performance of our models also supports previous findings that histone modification marks can help in identifying lncRNAs and DNA regions that have the potentials of forming RNA:DNA triplex structures. In conclusion, we showed that different deep learning architectures can recognize lncRNAs and DNA that have the potentials of forming RNA:DNA triplex structures.

bioinformatics↗

bioRxiv2025

OmicsFUSION: A Pretrained Hyena-Based Framework for Encoding, Reconstruction, and Representation of the Omics and DNA Data

Recent advances in Natural Language Processing (NLP) have spurred the application of Large Language Models (LLMs) to bioinformatics, enabling innovative approaches to DNA sequence encoding. However, genomic function is not solely determined by the primary DNA sequence--it also depends on complex, multi-layered omics data. This auxiliary data is inherently sparse, structured across multiple tracks, and poses a challenge for traditional unimodal approaches. Simultaneously, many bioinformatics tasks demand a unified signal source that encapsulates this information, rather than requiring researchers to input each omics feature individually. To overcome these limitations, we introduce a novel framework for the integrated representation of DNA sequences and their associated omics data. Central to our approach is a large-scale, batch-structured omics dataset optimized for deep learning at scale. Our framework is built around four key novelties: (1) a new model architecture that jointly processes DNA and omics signals; (2) an extension of Masked Language Modeling (MLM) to omics tracks for effective data reconstruction; (3) single-nucleotide embeddings that fuse all input modalities; and (4) interval-level embeddings that summarize broader genomic regions. We release a collection of pretrained models capable of reconstructing, embedding, and generating unified representations of DNA enriched with functional omics context. Codehttps://github.com/aaai-2025-submission/anomymous_submission_aaai2025 Model and Datasetshttps://drive.google.com/drive/folders/10hcuq4rTCr8bKQFqkpz7R91MGtKvR0Va?usp=sharing

bioinformatics↗

bioRxiv2025

OmiXAI: An Ensemble XAI Pipeline for Interpretable Deep Learning in Omics Data

Deep learning methods have become methods of choice in the analysis of genomic data. The performance of deep learning models depends on the information available for training. A growing trend in deep learning applications involves leveraging multi-omics data--spanning genomics, transcriptomics, epigenomics, proteomics, metabolomics, and other domains. When a deep learning model trained on omics data achieves high performance, the important question is to define factors that contribute to models predictive power. Explainable AI (XAI) methods can be categorized as model-aware and model-agnostic. Model-agnostic approaches, which rely on combinatorial feature perturbations to assess impact, are often computationally prohibitive for deep learning models. To address this, we developed OmiXAI, a pipeline integrating ensemble model-aware XAI methods. Our framework incorporates gradient-based techniques--including Integrated Gradients, InputXGradients, Guided Backpropagation, and Deconvolution (for CNNs and GNNs)--as well as Saliency Maps and GNNExplainer (specifically for GNNs). We evaluated OmiXAI on Z-DNA prediction using multi-omics features, demonstrating its efficacy through feature importance analysis and benchmarking of XAI methods. Notably, OmiXAI enabled feature engineering, reducing the critical feature set from almost 2,000 to just 50. Its modular design allows seamless integration of additional attribution methods, ensuring adaptability beyond omics to diverse problem domains. While testing the ensemble approach we benchmarked individual XAI methods and discuss their drawbacks and limitations. OmiXAI is freely available at https://github.com/aameliig/OmiXAI.

bioinformatics↗

bioRxiv2024

Kolmogorov-Arnold Networks for Genomic Tasks

Kolmogorov-Arnold Networks (KANs) emerged as a promising alternative for multilayer perceptrons in dense fully connected networks. Multiple attempts have been made to integrate KANs into various deep learning architectures in the domains of computer vision and natural language processing. Integrating KANs into deep learning models for genomic tasks has not been explored. Here, we tested linear KANs (LKANs) and convolutional KANs (CKANs) as replacement for MLP in baseline deep learning architectures for classification and generation of genomic sequences. We used three genomic benchmark datasets: Genomic Benchmarks, Genome Understanding Evaluation, and Flipon Benchmark. We demonstrated that LKANs outperformed both baseline and CK-ANs on almost all datasets. CKANs can achieve comparable results but struggle with scaling over large number of parameters. Ablation analysis demonstrated that the number of KAN layers correlates with the model performance. Overall, linear KANs show promising results in improving the performance of deep learning models with relatively small number of parameters. Unleashing KAN potential in different SOTA deep learning architectures currently used in genomics requires further research.

bioinformatics↗

bioRxiv2024

Analysis of live cell data with G-DNABERT supports a role for G-quadruplexes in chromatin looping

Alternative DNA conformation formed by sequences called flipons potentially alter the readout of genetic information by directing the shape-specific assembly of complexes on DNA The biological roles of G-quadruplexes formed by motifs rich in guanosine repeats have been investigated experimentally using many different methodologies including G4-seq, G4 ChIP-seq, permanganate nuclease footprinting (KEx), KAS-seq, CUT&Tag with varying degrees of overlap between the results. Here we trained large language model DNABERT on existing data generated by KEx, a rapid chemical footprinting technique performed on live, intact cells using potassium permanganate. The snapshot of flipon state when combined with results from other in vitro methods that are performed on permeabilized cells, allows a high confidence mapping of G-flipons to proximal enhancer and promoter sequences. Using G4-DNABERT predictions,with data from ENdb, Zoonomia cCREs and single cell G4 CUT&Tag experiments, we found support for a model where G4-quadruplexes regulate gene expression through chromatin loop formation.

bioinformatics↗

bioRxiv2024

Generative Models for Prediction of Non-B DNA Structures

MotivationDeep learning methods have been successfully applied to the tasks of predicting non-B DNA structures, however model performance depends on the availability of experimental data for training. Experimental technologies for non-B DNA structure detection are limited to the subsets that are active at the time of an experiment and cannot detect entire functional set of elements. Recently deep generative models demonstrated promising results in data augmentation approach improving classifier performance trained on augmented real and generated data. Here we aimed at testing performance of diffusion models in comparison to other generative models and explore the data augmentation approach for the task of non-B DNA structure prediction. ResultsWe tested denoising diffusion probabilistic and implicit models (DDPM and DDIM), Wasserstein generative adversarial network (WGAN) and vector quantised variational autoencoder (VQ-VAE) for the task of improving detection of Z-DNA, G-quadruplexes and H-DNA. We showed that data augmentation increased the quality of classifiers with diffusion models being the best for Z-DNA and H-DNA while WGAN worked better for G4s. Diffusion models are the best in diversity for all types of non-B DNA structures, WGAN produced the best novelty for G-quadruplexes and H-DNA. Since diffusion models require substantial resources, we showed that distillation technique can significantly enhance sampling in training diffusion models. When considering three criteria -quality of generated samples, sampling speed, and diversity, we conclude that trade-off is possible between generative diffusion model and other architectures such as WGAN and VQ-VAE. AvailabilityThe code with conducted experiments is freely available at https://github.com/powidla/nonB-DNA-structures-generation. Contactmpoptsova@hse.ru Supplementary informationSupplementary data are available at Journal Name online.

bioinformatics↗

bioRxiv2023

Z-Flipons conserved between human and mouse are associated with increased transcription initiation rates

A long-standing question concerns the role of Z-DNA in transcription. Here we use a deep learning approach based on the published DeepZ algorithm that predicts Z-flipons based on DNA sequence, structural properties of nucleotides and omics data. We examined Z-flipons that are conserved between human and mouse genomes after generating whole-genome Z-flipons maps by training DeepZ on ChIP-seq Z-DNA data, then overlapping the results with a common set of omics data features. We revealed similar pattern of transcription factors and histone marks associated with conserved Z-flipons, showing enrichment for transcription regulation coupled with chromatin organization. 15% and 7% of conserved Z-flipons fell in alternative and bidirectional promoters. We found that conserved Z-flipons in CpG-promoters are associated with increased transcription initiation rates. Our findings empower further experimental explorations to examine how the flip to Z-DNA alters the readout of genetic information by facilitating the transition of one epigenetic state to another.

bioinformatics↗

bioRxiv2023

Z-Flipon Variants reveal the many roles of Z-DNA and Z-RNA in health and disease

Identifying roles for Z-flipons remains challenging given their dynamic nature. Here we perform genome-wide interrogation with the DNABERT transformer algorithm trained on experimentally identified Z-DNA sequences. We show Z-flipons are enriched in promoters and telomeres and overlap quantitative trait loci for RNA expression, RNA editing, splicing and disease associated variants. Surprisingly, many effects are mediated through Z-RNA formation. We describe Z-RNA motifs present in SCARF2, SMAD1 and CACNA1 transcripts and others in non-coding RNAs. We also provide evidence for another Z-RNA motif that likely enables an adaptive anti-viral intracellular defense through alternative splicing of KRAB domain zinc finger proteins. An analysis of OMIM and gnomAD predicted loss-of-function datasets reveals an overlap of predicted and experimentally validated Z-flipons with disease causing variants in 8.6% and 2.9% of mendelian disease genes respectively, with frameshift variants present in 22% of cases. The work greatly extends the number of phenotypes mapped to Z-flipon variants.

genomics↗

bioRxiv2022

Graph Neural Networks for Z-DNA prediction in Genomes

Deep learning methods have been successfully applied to the tasks of predicting functional genomic elements such as histone marks, transcriptions factor binding sites, non-B DNA structures, and regulatory variants. Initially convolutional neural networks (CNN) and recurrent neural networks (RNN) or hybrid CNN-RNN models appeared to be the methods of choice for genomic studies. With the advance of machine learning algorithms other deep learning architectures started to outperform CNN and RNN in various applications. Thus, graph neural network (GNN) applications improved the prediction of drug effects, disease associations, protein-protein interactions, protein structures and their functions. The performance of GNN is yet to be fully explored in genomics. Earlier we developed DeepZ approach in which deep learning model is trained on information both from sequence and omics data. Initially this approach was implemented with CNN and RNN but is not limited to these classes of neural networks. In this study we implemented the DeepZ approach by substituting RNN with GNN. We tested three different GNN architectures - Graph Convolutional Network (GCN), Graph Attention Network (GAT) and inductive representation learning network GraphSAGE. The GNN models outperformed current state-of the art RNN model from initial DeepZ realization. Graph SAGE showed the best performance for the small training set of human Z-DNA ChIP-seq data while Graph Convolutional Network was superior for specific curaxin-induced mouse Z-DNA data that was recently reported. Our results show the potential of GNN applications for the task of predicting genomic functional elements based on DNA sequence and omics data. Availability and implementationThe code is freely available at https://github.com/MrARVO/GraphZ.

bioinformatics↗

bioRxiv2022

A role for Flipons and miRNAs in Promoter Specification during Development?

The classical view of gene regulation is based on prokaryotic models and the operon concept with protein-based transcription factors controlling the expression of metabolic pathways essential for bacterial adaptations in response to environmental changes. A new view for establishing cell identity is emerging in eukaryotes where RNA-based pathways provide the framework for the readout of genomic information. Another perspective poses that alternative DNA structures encoded by flipons enable switching of cellular responses from one state to another. Here we provide evidence that these RNA and DNA mechanisms are deeply connected. We present data supporting a model where flipons open up binding sites for microRNAs (miRNAs), leading to the establishment of bivalent promoters early in development whose location structures lineage-specific events. These outcomes are potentially influenced by ovarian and spermatozoan miRNAs, transmissions with evident evolutionary ramifications. The data supports a new perspective on genetic regulation, one in which the genome provides a canvas framed by flipons, sketched with miRNAs and embellished by proteins.

genetics↗

bioRxiv2022

Z-RNA and the flipside of the SARS Nsp13 helicase

We present evidence that the severe acute respiratory syndrome coronavirus (SARS) non-structural protein 13 (Nsp13) modulates the Z-RNA dependent regulated cell death pathways [1]. We show that Z-prone sequences (called flipons [2]) exist in coronavirus and provide a signature (Z-sig) that enables identification of the animal viruses from which the human pathogens arose. We also identify a potential RIP Homology Interaction Motif (RHIM) in the helicase Nsp13 that resembles those present in proteins that initiate Z-RNA-dependent cell death through interactions with the Z-RNA sensor protein ZBP1. These two observations allow us to suggest a model in which Nsp13 down regulates Z-RNA activated innate immunity by two distinct mechanisms. The first involves a novel ATP-independent Z-flipon helicase (flipase) activity in Nsp13 that differs from that of canonical A-RNA helicases. This flipase prevents formation of Z-RNAs that would otherwise activate cell death pathways. The second mechanism likely inhibits the interactions between ZBP1 and the Receptor Interacting Proteins Kinases RIPK1 and RIPK3 by targeting their RHIM domains. Together the described Nsp13 RHIM and flipase activities have the potential to alter the host response to coronaviruses and impact the design of drugs targeting the Nsp13 protein. The Z-sig and RHIM domains may provide a way of identifying previously uncharacterized viruses that are potentially pathogenic for humans.

microbiology↗