Biology subjects

Ahuja, G.

Publications and source records attributed to Ahuja, G..

At least 19 recordsLinked to original sources

bioRxiv2026

Three-dimensional mapping of human lung fibrosis identifies spatial coupling between tissue microarchitecture and maladaptive repair states

Human lung fibrosis progressively destroys three-dimensional tissue architecture, yet how the structural microenvironment shapes cellular states remains unclear. Integrating 3D X-ray histology with spatial omics, we generate a volumetric multimodal atlas of human lung fibrosis that enables direct interrogation of relationships between tissue microarchitecture and spatial molecular programmes. We identify collapse of the alveolar-capillary unit - the coupled loss of alveolar type 1 epithelial cells, capillary endothelial cells, and basement membrane integrity - as a structural transition associated with early fibrogenesis. Surrounding transitional zones exhibit aberrant epithelial differentiation and proteolytic matrix remodelling, with epithelium shifting toward bronchiolised, airway-like states. Stromal activation converges on spatially restricted fibrogenic cores, sites of maximal collagen maturation and matrix stabilisation that function as signalling hubs within surrounding perifocal and vascular microenvironments. Molecular signatures of atlas-defined tissue states were associated with lung function and mortality across independent patient cohorts. Together, these data support a model in which fibrosis emerges as a spatial tissue-state process, with alveolar-capillary collapse redirecting repair towards mechanically stabilised niches that reinforce pathological remodelling.

systems biology↗

bioRxiv2026

Scalable Computational Phenomics of Nuclear Morphology

Interpretable computational pathology is constrained by a trade-off between scalable learned representations and biologically explicit phenotypes. Here we present NucXplore, a nuclear phenomics framework that converts each hematoxylin-and-eosin (H&E)-stained nucleus into 129 explicitly defined features spanning morphology, chromatin, intensity, texture, color, and spatial context. A Rust-based implementation accelerates extraction 17.6-fold over the Python reference implementation. Across three histopathology cohorts, NucXplore outperformed classical descriptors and larger deep-learning embeddings while retaining feature attribution. Applied to human skin, hierarchical multiple-instance models resolved eight cellular compartments and decoded age and sun-exposure states, revealing a largely shared aging program with age and cell-specific exposure-dependent recalibration. Conditional flow matching and neural ODE integration transformed cross-sectional phenotypes into model-implied trajectories, uncovering a midlife velocity minimum, late-life re-acceleration, and cell-specific changes in trajectory magnitude and direction. Source-free domain adaptation enabled preliminary transfer of age-associated predictions to an independent hospital cohort. NucXplore thus unifies interpretable representation, high-performance computing, and dynamical modeling for cellular phenotyping from routine histology.

pathology↗

bioRxiv2026

SMARTIE: A Machine-Learning approach for investigating RBP-RNA interactions identified by Editing

RNA-binding proteins (RBPs) play important roles in gene regulation. RNA editing-based approaches, such as TRIBE and STAMP, have gained wider use for identifying RNA targets of RBPs. These methods offer advantages over crosslinking-based approaches in terms of experimental simplicity and in vivo applicability. However, data analysis methods for these approaches remain underdeveloped, limiting sensitivity, and unbiased target prioritization. To address these limitations, we introduce SMARTIE (Systematic Machine-learning Approach for RBP Targets Identified by Editing), a machine-learning-based framework. SMARTIE robustly identifies and ranks RBP target RNAs from editing data by integrating statistical tests with replicate-aware and confidence-weighted features. Reanalysis of multiple published TRIBE datasets demonstrates the effectiveness of SMARTIE. It recovers targets of RBPs like Ataxin-2, TDP-43, Hrp48, Thor, GPATCH8, dFMRP and NonA. Notably, a model trained on TRIBE data generalizes to STAMP datasets, suggesting that SMARTIE learns universal signatures of editing-based RBP targeting there by enabling more accurate inference for RBP-RNA interactions. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=97 SRC="FIGDIR/small/726004v1_ufig1.gif" ALT="Figure 1"> View larger version (34K): org.highwire.dtl.DTLVardef@8b77e5org.highwire.dtl.DTLVardef@6c9416org.highwire.dtl.DTLVardef@6e33a5org.highwire.dtl.DTLVardef@100b7b5_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2026

A multitask single-cell analysis framework with knowledge graph as a prior

Single-cell RNA sequencing has reshaped cancer biology by revealing the cellular and molecular diversity of the tumor microenvironment, yet most analysis pipelines still treat clustering, annotation, pathway scoring, metabolic inference, and cell-cell communication as separate tasks. Here, we present scKNIFE (Knowledge graph-based NMF for Inference of Functional Entities), a unified graph-regularized non-negative matrix factorization framework that jointly reconstructs single-cell expression profiles and infers activities of diverse biological entities, including pathways, cell-type markers, metabolic reactions and tasks, ligand-receptor interactions, and cell-state programs. By embedding a heterogeneous knowledge graph directly into the factorization objective, scKNIFE propagates information between biologically related entities through Laplacian regularization while sparsity constraints preserve interpretability. The integrated prior graph spans 47,274 nodes and 506,620 edges, assembled from complementary resources including Gene Ontology, Reactome, Hallmark gene sets, Human-GEM, CellChat, PanglaoDB, Cytopus, and Tabula Muris-derived annotations. Across multiple cancer-focused single-cell datasets, scKNIFE yields competitive to leading performance for clustering and cell-type annotation, while also recovering biologically coherent metabolic programs that agree with dedicated metabolism-focused methods and capture therapy-response-associated states. In addition, the framework supports downstream inference of cell-type-specific biological activities from a single latent representation. Together, these results establish scKNIFE as a modular and extensible framework for end-to-end biological interpretation of single-cell cancer transcriptomics

cancer biology↗

bioRxiv2026

Deterministic retrieval recovers biomedical associations lost by language models

Biomedical research increasingly relies on expert-curated databases to connect diseases, genes, variants, phenotypes, pathways and therapeutics. Incomplete or irreproducible retrieval can distort interpretation, mechanistic inference, variant assessment or therapeutic prioritization despite an unchanged evidence base. Agentic natural-language-to-SQL (NL2SQL) systems and Model Context Protocol-enabled agents can autonomously retrieve evidence from biomedical databases, including Open Targets and the Highly Confident Drug-Target Database, in response to natural-language questions. However, they may truncate large result sets, miss records when terminology differs from database vocabulary and return different outputs across runs. We systematically quantify these failures and introduce BioChirp, a goal-directed autonomous system for reliable retrieval from structured biomedical databases, built on interpretation-execution separation. A coordinated language-model layer interprets intent and resolves database-field assignments, after which entity resolution maps user terms to database terminology. A deterministic Steiner-tree planner and executor then retrieve matching records without further language-model involvement. Across curated biomedical databases spanning more than one million associations, BioChirp retrieved thousands of records for exhaustive queries, whereas baseline agents returned at most a few hundred or failed. Across 70 queries, five runs and three BioChirp database backends, BioChirp achieved a median cross-run Jaccard similarity of 1.0. Across 910 database-question evaluations derived from expert-written BioASQ questions and spanning ten databases, BioChirp retrieved at least one database record in 85.4% of cases and produced a fully correct answer in 51.1% of cases, compared with 44.6% and 17.4% for agentic NL2SQL. BioChirp is publicly available at https://biochirp.iiitd.edu.in.

bioinformatics↗

bioRxiv2026

Divergent phenotypic and functional roles of human T follicular helper cells from infancy to adulthood

Antibody responses to T-dependent antigens are suboptimal in young children, yet the evolution of T follicular helper cell (Tfh) function across the human lifespan remains poorly defined. Using human tonsils, a physiologically relevant and abundant source of Tfh, we investigated age-associated differences in their repertoire and functional programs. Pediatric tonsils were enriched for cytokine-expressing Tfh subsets with increased clonal diversity and phenotypic plasticity. However, in response to influenza antigens, they exhibited reduced Th1 polarization, diminished IL-21 production, and limited B cell help. Across ages, high neutralizing flu antibody responses were associated with robust Tfh1 activation, which was ICOS dependent in adults but not in children. Interestingly, Tfh depletion strategies revealed enhanced Tfh differentiation from distinct precursors in pediatric donors, yet antibody responses during early life were less reliant on Tfh help. Together, these findings define developmentally programmed differences in Tfh differentiation and function with implications for pediatric vaccine design.

immunology↗

bioRxiv2026

A tonsil organoid model reveals Epstein-Barr virus infected germinal center B cell states during primary infection

Epstein-Barr virus (EBV) colonizes secondary lymphoid tissues to establish persistent infection and is strongly associated with malignancy and autoimmunity. Our understanding of EBV infection biology is hindered by a lack of models that capture infected B cell activity in the lymphoid tissue microenvironment. We therefore developed an EBV human tonsil organoid model to evaluate key B cell states and antiviral responses, including after primary infection. EBV promoted B cell differentiation into germinal center (GC)-like phenotypes and transcriptomic analyses highlighted numerous B cell transcriptional programs unique to EBV-infected cells. B cell receptor repertoire analysis revealed that most EBV-infected B cells underwent class switching but only rarely participated in somatic hypermutation. CD4 T cells, highly activated by organoid infection, limited EBV+ B cell outgrowth. Our findings demonstrate human tonsil organoids as a physiologically relevant model to investigate key aspects of EBV immunity and pathogenesis.

immunology↗

bioRxiv2025

Chemical Dice Integrator (CDI): A Scalable Framework for Multimodal Molecular Representation Learning

The machine learning landscape for molecular property prediction is fragmented, with numerous Featurizers each capturing a narrow, specialized view of chemical structure. This heterogeneity forces a suboptimal choice of representation a priori, limiting model generalizability. We introduce the Chemical Dice Integrator (CDI), a hierarchical framework that unifies six orthogonal molecular representations, physicochemical (Mordred), topological (GROVER), visual (ImageMol), biological (Signaturizer), quantum-mechanical (MOPAC), and linguistic (ChemBERTa), into a single, coherent embedding. The framework consists of CDI-Basic, a two-tiered autoencoder that fuses these modalities, and CDI-Generalised, a Mamba State-Space Model (SSM) that learns a direct, efficient map from SMILES strings to the unified embedding space. Extensive benchmarking across 23 classification (171 tasks) and 10 regression datasets demonstrates that CDI embeddings consistently achieve superior predictive performance compared to individual Featurizers and standard feature aggregation methods. The CDI-Generalised model achieves this performance with exceptional computational efficiency, outperforming deep learning Featurizers in terms of speed and resource overhead. Furthermore, we demonstrate that the CDI embedding is chemically intuitive, allowing for the sensitive distinction of nuanced structural variants, such as chiral enantiomers and kekulized SMILES forms. By bridging multimodal chemical intelligence with scalable, sequence-based inference, CDI offers a strong foundation for molecular machine learning.

bioinformatics↗

bioRxiv2025

Multi-agent AI enables evidence-based cell annotation in single-cell transcriptomics

Cell type annotation remains a critical bottleneck, with current methods often inaccurate and requiring extensive manual validation, particularly in disease contexts. While large language models (LLMs) show promise, they can be unreliable due to hallucinations. We developed CyteType, a multi-agent framework that generates competing hypotheses grounded in full expression data and study context, validates against external databases, and iteratively self-evaluates. Comprehensive benchmarking demonstrates that CyteType substantially outperforms reference-based and LLM-based methods, with self-generated confidence scores reliably identifying trustworthy annotations. CyteType transforms cell type annotation from label assignment into evidence-grounded biological discovery. Python (AnnData compatible): https://github.com/NygenAnalytics/CyteType R (Seurat compatible): https://github.com/NygenAnalytics/CyteTypeR

bioinformatics↗

bioRxiv2025

Mechanism-aware inference of response to targeted cancer therapies

Targeted therapies like small-molecule inhibitors often work by blocking proteins that cancer cells rely on for survival. Omics based modeling of drug sensitivity alone lack mechanistic grounding. We propose FORGE (Factorization Of Response and Gene Essentiality) a simple yet powerful joint matrix factorization framework that co-models drug response and target gene essentiality, enabling the stratification of promising treatment groups for targeted therapy consideration. FORGE also provides Benefit Score -- a predictive score that estimates treatment efficacy from basal gene expression profiles. We validated the predictive performance of FORGE across multiple targeted therapies, including Erlotinib (EGFR inhibitor) and Daporinad (NAMPT inhibitor). Our meta-analysis of large scale in-vitro studies underscores FORGEs ability to identify common determinants of drug vulnerabilities and target gene essentiality. Such convergences were not observed when treatment vulnerabilities and gene essentialities were modeled independently. We also demonstrated the universality of Erlotinib Benefit Scores by transferring transformations learned from high-throughput drug response studies across other published datasets, including the TAHOE-100M single-cell perturbation atlas and patient-derived xenograft studies. FORGE successfully identified key regulators within the molecular pathways targeted by these therapies, reinforcing its potential for mechanistically grounded treatment stratification.

cancer biology↗

bioRxiv2025

SynGlue: AI-Driven Designer for Clinically Actionable Multi-Target Therapeutics

The rational design of protein degraders, such as proteolysis-targeting chimeras (PROTACs), requires the simultaneous optimization of multiple molecular properties, a complex challenge that limits efficient discovery. Here, we introduce SynGlue, a generative artificial intelligence (AI) framework that addresses this challenge through two core modules: data-driven, leveraging large-scale protein-ligand intelligence, and structure-guided, for physics-aware molecular design. SynGlue harness MagnetDB, a curated database of 6.37 million experimental protein-ligand interactions, and couples it with deep learning models that quantitatively predict degradation potency (DC50), maximal degradation (Dmax), and guide ternary-complex-compatible linker design. Benchmarked against 6,935 compounds, SynGlue demonstrates superior performance in relevant pharmacology prediction. To validate SynGlue, we engineered degraders for BRD4 and GSPT1. Our data-driven design for BRD4 yielded compounds with novel warhead scaffolds (<50% warhead similarity with known PROTACs), which proved to be potent degraders in vitro (DC50 = 0.19 nM) and efficacious in vivo in mouse models. Independently, our structure-guided de novo design for GSPT1 produced ultrapotent degraders (DC50 {approx} 0.0011 M) that are also effective both in vitro and in vivo, uncovering a new oncogenic dependency. By unifying data-driven and physics-aware design, SynGlue establishes a generalizable AI framework for the rapid development of clinically relevant protein degraders, with principled extension to other multi-target modalities.

bioinformatics↗

bioRxiv2025

MutAIverse: An AI-Powered, Mechanism-backed Platform for Discovering Novel DNA Adducts and their precursor Genotoxins

Genotoxin exposure leads to DNA adduct formation, potentially causing mutations if unrepaired. Current DNA adductomics platforms or analytical workflows are limited by incomplete spectral libraries, reliance on experimentally validated adducts, limited cellular contexts, and inefficient computational methodologies. We introduce MutAIverse, an advanced AI-driven DNA adductomics analysis platform that overcomes these limitations by leveraging intracellular mechanistic modeling of genotoxin bioactivation to construct a comprehensive DNA adduct library and offers advanced spectral mapping. MutAIverse integrates experimentally validated and chemically valid putative DNA adducts, enabling interpretable retrograde tracking to parental genotoxins. Validation of MutAIverse against experimental MS/MS datasets demonstrated the detection of both known and novel adducts. Furthermore, application to in-house generated DNA adductomics data from tissue biopsies of smokeless tobacco-induced head and neck cancer patients revealed selective enrichment of both novel and known DNA adducts; a subset of them was validated using MS/MS analysis. Collectively, MutAIverse provides a robust, end-to-end, and interpretable platform for advanced DNA adductomics analysis.

bioinformatics↗

bioRxiv2024

RCoxNet: deep learning framework for enhanced cancer survival prediction integrating random walk with restart with mutation and clinical data

Cancer poses a significant global health challenge, characterized by a complex disease progression and disrupted growth regulation. A thorough understanding of cellular and molecular biological mechanisms is essential for developing novel treatments and improving the accuracy of patient survival predictions. While prior studies have leveraged gene expression and clinical data to forecast survival outcomes through current machine learning and deep learning approaches, gene mutation data--despite being a widely recognized metric--has rarely been incorporated due to its limited information, inadequate representation of gene relationships, and data sparsity, which negatively affects the robustness, effectiveness, and interpretability of current survival analysis approaches. To overcome the challenges of mutation data sparsity, we propose RCoxNet, a novel deep learning neural network framework that integrates the Random Walk with Restart (RWR) algorithm with a deep learning Cox Proportional Hazards model. By applying this framework to mutation data from cBioportal, our model achieved an average concordance index of 0.62 {+/-} 0.05 across four cancer types, outperforming existing deep neural network models. Additionally, we identified clinical features critical for differentiating between predicted high- and low-risk patients, with the relevance of these features being partially supported by previous studies.

bioinformatics↗

bioRxiv2024

Tissue determinants of the human T cell receptor repertoire.

98% of T cells reside in tissues, yet nearly all human T cell analyses are performed from peripheral blood. We single-cell sequenced 5.7 million T cells from ten donors autologous blood and tonsils and sought to answer key questions about T cell receptor biology previously unanswerable by smaller-scale experiments. We identified distinct clonal expansions and distributions in blood compared to tonsils, with surprisingly low (1-7%) clonal sharing. These few shared clones exhibited divergent phenotypes across bodily sites. Analysis of antigen-specific CD8 T cells revealed location as a main determinant of frequency, phenotype, and immunodominance. Finally, diversity estimates from the tissue recalibrates current repertoire diversity estimates, and we provide a refined estimate of whole-body repertoire. Given the tissue-restricted nature of T cell phenotypes, functions, differentiation, and clonality revealed by this dataset, we conclude that tissue analyses are crucial for accurate repertoire analysis and monitoring changes after perturbing therapies.

immunology↗

bioRxiv2024

Selective promotion of oligonucleotides in the course of evolution

Evolution shapes the structure and content of genomes, yet the contribution of local sequence composition to variant selection remains poorly understood. While traditional models emphasize protein function or cross-species conservation, we propose that intra-genomic patterns of oligonucleotide (k-mer) frequencies also reflect selective forces. To explore this, we developed kGain score, a metric that quantifies the frequency shift of a k-mer upon single-nucleotide substitution, using the surrounding genomic context as a baseline. We hypothesize that variants arising in high-kGain contexts are more likely to persist due to evolutionary favorability. We validated this hypothesis across multiple systems. In E. coli and S. cerevisiae long-term evolution experiments, we found that fixed, essential, and parallel mutations consistently show elevated kGain scores. This trend held in SARS-CoV-2 variants of concern and in an in-house antibiotic adaptation experiment, where a high-kGain fusA Y515N mutation conferred resistance and maintained fitness when overexpressed, demonstrating a causal link between kGain and adaptive potential. To enable cross-species generalization, we trained a transformer-based neural network regressor on LTEE-derived mutations to predict kGain from sequence alone. The model achieved high correlation in held-out in-domain data (Pearson r = 0.81) and accurately predicted kGain trends in out-of-domain data (Pearson r = 0.82), demonstrating that k-mer-based sequence constraints learned from one genome can be effectively transferred to others. Together, our results establish kGain as a biologically meaningful, scalable metric for probing within-genome sequence selection, offering a complementary lens to existing conservation-based frameworks for understanding evolutionary fitness and variant persistence.

evolutionary biology↗

bioRxiv2023

Artificial Intelligence uncovers Evolutionarily Conserved Intracellular Allosteric Modulators of GPCR-Gα interface.

Endogenous intracellular allosteric modulators of GPCRs remain largely unexplored, with limited binding and phenotype data available. This gap arises from the lack of robust computational methods for unbiased cavity identification, cavity-specific ligand design, synthesis, and validation across GPCR topology. Here, we developed Gcoupler, an AI-driven generalized computational toolkit that leverages an integrative approach combining de novo ligand design, statistical methods, Graph Neural Networks, and Bioactivity-based ligand prioritization for rationally predicting high-affinity ligands. Using Gcoupler, we interrogated intracellular metabolites that target and regulate the GPCR-G interface (Ste2p-Gpa1p), affecting pheromone-induced programmed cell death in yeast. Our computational analysis, complemented by experimental validations, including genetic screening, multi-omics, site-directed mutagenesis, biochemical assays, and physiological readouts, identified endogenous hydrophobic metabolites, notably sterols, as direct intracellular allosteric modulators of Ste2p. Molecular simulations coupled with biochemical signaling assessment in site-directed Ste2p mutants further confirmed that metabolites binding to GPCR-G obstruct downstream signaling, possibly via a cohesive effect. Finally, by utilizing isoproterenol-induced, GPCR-mediated human and neonatal rat cardiac hypertrophy models, we observed that elevated metabolite levels attenuate hypertrophic response, reinforcing the evolutionary relevance of this mechanism.

bioinformatics↗

bioRxiv2022

Latent representation of single-cell transcriptomes enables algebraic operations on cellular phenotypes

Single-cell RNA-sequencing (scRNA-seq) coupled with robust computational analysis facilitates the characterization of phenotypic heterogeneity within tumors. Current scRNA-seq analysis pipelines are capable of identifying a myriad of malignant and non-malignant cell subtypes from single-cell profiling of tumors. However, given the extent of intra-tumoral heterogeneity, it is challenging to assess the risk associated with individual cell subpopulations, primarily due to the complexity of the cancer phenotype space and the lack of clinical annotations associated with tumor scRNA-seq studies. To this end, we introduce SCellBOW, a scRNA-seq analysis framework inspired by document embedding techniques from the domain of Natural Language Processing (NLP). SCellBOW is a novel computational approach that facilitates effective identification and high-quality visualization of single-cell subpopulations. We compared SCellBOW with existing best practice methods for its ability to precisely represent phenotypically divergent cell types across multiple scRNA-seq datasets, including our in-house generated human splenocyte and matched peripheral blood mononuclear cell (PBMC) dataset. For tumor cells, SCellBOW estimates the relative risk associated with each cluster and stratifies them based on their aggressiveness. This is achieved by simulating how the presence or absence of a specific cell subpopulation influences disease prognosis. Using SCellBOW, we identified a hitherto unknown and pervasive AR-/NElow (androgen-receptor-negative, neuroendocrine-low) malignant subpopulation in metastatic prostate cancer with conspicuously high aggressiveness. Overall, the risk-stratification capabilities of SCellBOW hold promise for formulating tailored therapeutic interventions by identifying clinically relevant tumor subpopulations and their impact on prognosis.

bioinformatics↗

bioRxiv2022

Simultaneous visualization of cells and marker genes from scRNA-seq studies

The complexity of scRNA-sequencing datasets highlights the urgent need for enhanced clustering and visualization methods. Here, we propose Stardust, an iterative, force-directed graph layouting algorithm that enables the simultaneous embedding of cells and marker genes. Stardust, for the first time, allows a single-stop visualization of cells and marker genes on a single 2D map. While Stardust provides its own visualization pipeline, it can be plugged in with state-of-the-art methods such as Uniform Manifold Approximation and Projection (UMAP) and t-Distributed Stochastic Neighbor Embedding (tSNE). We benchmarked Stardust against popular visualization and clustering tools on both scRNA-seq and spatial transcriptomics datasets. In all cases, Stardust performs competitively in identifying and visualizing cell types in an accurate and spatially coherent manner.

bioinformatics↗