Biology subjects

Bhati, U.

Publications and source records attributed to Bhati, U..

5 recordsLinked to original sources

bioRxiv2026

GANGE: Achieving Sequencing Without Sequencing With Diffusion Guided Generative Genomic Transformer

The genome of a species is its book of life, but opening that book remains a costly affair due to the limitations the existing sequencing technologies pose. Short reads sequencers struggle to capture long and complex genomes, though have high fidelity rate. To counter that long reads from IIIrd generation sequencers are used, which are full of indel errors. Thus, reads from both approaches are collectively used with very high coverage, making the sequencing projects unreasonably high of cost and unapproachable to majority. Here we present a first of its kind generative deep-learning system, GANGE, which not just recovers the correct sequence with high accuracy from indel prone ONT reads at manifold lesser coverage but also extends it by 4kb, achieving sequencing without sequencing, horizontally as well as vertically while maintaining >92% accuracy consistently. This all makes it possible to drastically pull down sequencing project cost. GANGE was tested across A. thaliana, O. sativa genomes and Human chromosome 1 where it delivered outstanding assembly performance. Besides this, it was also used to accurately generate 2kb upstream promoters of all the genes from 12 different species, demonstrating that one can now also take up regulomics research just using RNA data alone when genome sequence is not available. With this all, GANGE brings a democratic turning point in the area of genomics and sequencing research.

bioinformatics↗

bioRxiv2026

ProMaya: a hierarchical universal Deep Learning framework for accurate and interpretable Protein-Protein interaction identification

Protein-protein interactions (PPIs) are molecular lego which define the physical states of cells. Accurately identifying PPIs remains challenging due to the interplay of several factors ranging from electrostatic to molecular geometry, topology, and physics. Existing computational approaches capture only fragments of this orchestra, limiting their generalizability across protein families and interaction types. Here, we present ProMaya, a hierarchical multi-scale Graph-transformer framework that integrates 3D atomic geometry, electronic distribution, residue-level structure and disorder, surface mass-density signatures, and large protein language-model embeddings of interacting proteins. Highly comprehensively benchmarked across nine species and 47 GB experimentally validated data, ProMaya achieved consistently >95% average accuracy, outperforming state-of-the-art tools by >12%. As driven by its explainability, the first time introduced atomic and protein language information dramatically boosted it to an outstanding level for PPI discovery in any species, potent to even bypass costly experiments. ProMaya system is freely accessible at https://scbb.ihbt.res.in/ProMaya/

bioinformatics↗

bioRxiv2025

Decoding stress specific transcriptional regulation by causality aware Graph-Transformer deep learning

Cells respond to environmental stimuli through transcriptional reprogramming orchestrated by transcription factors (TFs), which interpret cis-regulatory DNA sequences to determine the timing and location of gene expression. The diversification of TFs and their interactions with cis-regulatory elements (CREs) underpins plant adaptation to stress through the formation of gene regulatory networks (GRNs). However, deciphering condition-specific GRNs and identifying transcription factor binding motifs (TFBMs) for spatio-temporal gene expression remain major challenges in plant biology. To decipher the conditional networks governing TF-Target gene interactions, we developed CTF-BIND, a novel computational framework designed to reason about the spatio-temporal dynamics of TF activity. Leveraging over [~]23TB of multi-omics data (ChIP-seq, RNA-seq, and protein-protein interaction data), we constructed Bayesian causal networks capable of explaining TF activity across diverse conditions. These networks, validated against extensive experimental data, were then integrated into a Graph Transformer deep learning system. This system uses expression information of network components to quantitatively determine TF activity levels. Models were developed for 110 abiotic stress-related TFs, enabling accurate condition-specific detection of TF binding directly from RNA-seq data, eliminating the need for separate ChIP-seq experiments. CTF-BIND achieved a high average accuracy of [~]93% when tested against experimentally established data from various conditions. It is implemented as an interactive, open-access web server, it not only provides TF binding profiles but also facilitates downstream functional analysis. Furthermore, we developed CTF-BIND-DB, (https://hichicob.ihbt.res.in/ctfbind/) a database capturing dynamic shifts in regulatory pathways, providing information on TGs, network ontology, and binding motifs. CTF-BIND and CTF-BIND-DB represent a transformative approach for understanding and determining TF activity in plant stress responses, offering a powerful tool for crop improvement and bypassing the limitations of traditional methods and extensive experimental validation.

bioinformatics↗

bioRxiv2024

PTF-Vac: Ab-initio discovery of plant transcription factors binding sites using deep co-learning encoders-decoders

Discovery of transcription factors (TFs) binding sites (TFBS) and their motifs in plants pose significant challenges due to high cross-species variability. The interaction between TFs and their binding sites is highly specific and context dependent. Most of the existing TFBS finding tools are not accurate enough to discover these binding sites in plants. They fail to capture the cross-species variability, interdependence between TF structure and its TFBS, and context specificity of binding. Since they are coupled to predefined TF specific model/matrix, they are highly vulnerable towards the volume and quality of data provided to build the motifs. All these software make a presumption that the user input would be specific to any particular TF which renders them of very limited use for practical applications like genomic annotations of newly sequenced species. Here, we report an explainable Deep Encoders-Decoders generative system, PTF-V[a]c, founded on a universal model of deep co-learning on variability in binding sites and TF structure, PTFSpot, making it completely free from the bottlenecks mentioned above. It has successfully decoupled the process of TFBS discovery from the prior step of motif finding and requirement of TF specific motif models. Due to the universal model for TF:DNA interactions as its guide, it can discover the binding motifs in total independence from data volume, species and TF specific models. In a comprehensive benchmarking study across a huge volume of experimental data, it has outperformed most advanced motif finding deep learning (DL) algorithms. With this all, PTF-V[a]c brings a completely new chapter in ab-initio TFBS discovery through generative AI. Short SummaryThe discovery of transcription factor binding sites (TFBS) in plants is challenging due to high variability across species and context-specific interactions. Traditional tools rely on predefined models and often fail in their cross-species applications. PTF-V[a]c has implemented generative a universal deep-learning model that decouples TFBS discovery from predefined motifs, enabling accurate, species-independent TFBS/motif identification while outperforming existing methods by huge leads.

bioinformatics↗

bioRxiv2023

Deep Co-learning on transcription factors and their binding sites attains impeccable universality in plants

Unlike animals, variability in transcription factors (TF) and their binding regions (TFBR) across the plants species is a major problem which most of the existing TFBR finding software fail to tackle, rendering them hardly of any use. This limitation has resulted into underdevelopment of plant regulatory research and rampant use of Arabidopsis like model species, generating misleading results. Here we report a revolutionary transformers based deep-learning approach, PTFSpot, which learns from TF structures and their binding regions co-variability to bring a universal TF-DNA interaction model to detect TFBR with complete freedom from TF and species specific models limitations. During a series of extensive benchmarking studies over multiple experimentally validated data, it not only outperformed the existing software by >30% lead, but also delivered consistently >90% accuracy even for those species and TF families which were never encountered during model building process. PTFSpot makes it possible now to accurately annotate TFBRs across any plant genome even in the total lack of any TF information, completely free from the bottlenecks of species and TF specific models.

bioinformatics↗