Biology subjects

Sethi, T.

Publications and source records attributed to Sethi, T..

6 recordsLinked to original sources

bioRxiv2026

Benchmarking Large Language Models for Assay-Aware Drug-Target Binding Affinity Prediction

Machine learning models for molecular discovery often train on bioactivity data from heterogeneous assays, yet assay context is rarely explicitly represented. We investigate whether natural-language assay descriptions provide predictive information beyond molecular and protein representations. We develop ACL-DTI (Assay-aware Contrastive Learning-based Drug Target Interaction) and MAMMAL-AsAP, an assay-context extension of MAMMAL-DTI. Across five triplet-disjoint random splits, assay-aware models achieve competitive or improved performance over assay-agnostic baselines. Among the evaluated models, MAMMAL-AsAP achieves the highest (R^2) of 0.7198 +/- 0.0085. Masking experiments show that ACL-DTI is robust to incomplete assay descriptions, exhibiting the smallest reductions in R^2 among the evaluated models when masking the target name, assay technique, or both: 0.0228 +/- 0.0064, 0.0406 +/- 0.0113, and 0.0620 +/- 0.0190, respectively. These results indicate that assay descriptions contain predictive information not fully captured by molecular structures and protein sequences, and reveal architecture-dependent differences in how this information is exploited.

bioinformatics↗

bioRxiv2026

Generative Augmentation Reveals Previously Overlooked Signals in Transcriptomic Datasets

Identifying robust gene expression signatures from transcriptomic studies with small sample sizes remains one of the most persistent challenges in computational biology. Gene expression datasets have thousands of features but only a handful of biological samples. This presents the classic p >> n imbalance, which limits statistical power and makes it difficult to discover reliable biomarkers. In imaging, generative models such as GANs, VAEs, and diffusion models have demonstrated promising applications in data augmentation, but their usefulness for omics data has not been systematically tested. More importantly, no existing framework integrates synthetic data generation, stability-aware signature discovery, and multi-source biological validation into a single pipeline. In this work, we present GeneLift, with the hypothesis that a computational pipeline of generative data augmentation, stability testing, and evaluating biological evidence will aid novel gene-signature discovery in small-cohort transcriptomic studies. We tested this hypothesis across 36 microarray datasets covering five diseases: sepsis, breast cancer, ovarian cancer, tuberculosis, and diabetes. A component-wise testing of GeneLift revealed that Gaussian Mixture Models (GMMs) outperformed deep generative approaches and faithfully reproduced gene-level distributions. By a novel approach of titrating the level of augmentation, we identified biologically meaningful gene candidates that did not appear in the original, underpowered analyses. We also developed BayesScore, a Bayesian posterior probability of gene-disease association computed from PubMed co-occurrence, which both recovers well-characterised disease genes missed by standard differential expression and surfaces candidates whose disease relevance was independently confirmed in subsequent publications, with lead times of up to 18 years between the source dataset and the first disease-specific citation. GeneLift is freely available at tavlab-iiitd/GeneLift. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=103 SRC="FIGDIR/small/720348v1_ufig1.gif" ALT="Figure 1"> View larger version (41K): org.highwire.dtl.DTLVardef@13c8aeeorg.highwire.dtl.DTLVardef@1c5277corg.highwire.dtl.DTLVardef@a35664org.highwire.dtl.DTLVardef@17cdc48_HPS_FORMAT_FIGEXP M_FIG C_FIG

genomics↗

bioRxiv2022

CD98 is critical for a conserved inflammatory response to diverse injury stimuli relevant to IPF exacerbations and COVID pneumonitis

Progressive fibrosing interstitial lung diseases (PFILDs) cause substantial morbidity and mortality. Antifibrotic agents slow progression, but most of the clinical need remains unmet. The archetypal PFILD is idiopathic pulmonary fibrosis (IPF). Chronic progression is driven by transforming growth factor (TGF-){beta}1 signalling. It is punctuated by inflammatory flares known as acute exacerbations (AE-IPF), which are associated with accelerated decline and high mortality. We hypothesized that acute injury responses underlying exacerbations and the mechanisms of chronic fibrosis overlap at the molecular level, via a cell surface assembly nucleated by galectin-3 that we term the gal-3-fibrosome. We focused upon a putative pro-inflammatory galectin-3 ligand, the CD98:integrin complex. Our data indicate CD98 and {beta}1-integrin co-localise with galectin-3 within epithelial cells in IPF lung tissue, and within 40 nm in human lung tissue treated with TGF-{beta}1 compared to controls. CD98 is required for interleukin (IL-)6 and IL-8 responses to biochemical and biophysical conditions mimicking stimuli of AE-IPF in vivo, ex vivo and in cells, and for an interstitial neutrophilic response in a mouse model. We demonstrate this pathway progresses via intracellular influx of Ca2+ mediated by TRPV4, and NF-{kappa}B activation, operating in positive feedback. Lastly we show the CD98- and galectin-3-dependence of IL-6 and IL-8 responses to the SARS-CoV-2 spike protein receptor binding domain and the conservation of this response pattern between lung epithelial cells and monocyte-derived macrophages. Taken together our findings identify CD98 as a key mediator of both pro-fibrotic and acute inflammatory responses in the lung with relevance to AE- and chronic progression of IPF, and the priming of fibrotic lungs for acute inflammatory responses. They similarly implicate CD98 and galectin-3 as mediators of COVID pneumonitis and worse outcomes in ILD patients with COVID.

cell biology↗

bioRxiv2021

Genomic Surveillance of COVID-19 Variants with Language Models and Machine Learning

The global efforts to control COVID-19 are threatened by the rapid emergence of novel SARS-CoV-2 variants that may display undesirable characteristics such as immune escape, increased transmissibility or pathogenicity. Early prediction for emergence of new strains with these features is critical for pandemic preparedness. We present Strainflow, a supervised and causally predictive model using unsupervised latent space features of SARS-CoV-2 genome sequences. Strainflow was trained and validated on 0.9 million sequences for the period December, 2019 to June, 2021 and the frozen model was prospectively validated from July, 2021 to December, 2021. Strainflow captured the rise in cases two months ahead of the Delta and Omicron surges in most countries including the prediction of a surge in India as early as beginning of November, 2021. Entropy analysis of Strainflow unsupervised embeddings clearly reveals the explore-exploit cycles in genomic feature-space, thus adding interpretability to the deep learning based model. We also conducted codon-level analysis of our model for interpretability and biological validity of our unsupervised features. Strainflow application is openly available as an interactive web-application for prospective genomic surveillance of COVID-19 across the globe.

bioinformatics↗

bioRxiv2021

Evaluating Sample Augmentation in Microarray Datasets with Generative Models: A Comparative Pipeline and Insights in Tuberculosis.

High throughput screening technologies have created a fundamental challenge for statistical and machine learning analyses, i.e., the curse of dimensionality. Gene expression data are a quintessential example, high dimensional in variables (Large P) and comparatively much smaller in samples (Small N). However, the large number of variables are not independent. This understanding is reflected in Systems Biology approaches to the transcriptome as a network of coordinated biological functioning or through principal Axes of variation underlying the gene expression. Recent advances in generative deep learning offers a new paradigm to tackle the curse of dimensionality by generating new data from the underlying latent space captured as a deep representation of the observed data. These have led to widespread applications of approaches such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), especially in domains where millions of data points exist, such as in computer vision and single cell data. Very few studies have focused on generative modeling of bulk transcriptomic data and microarrays, despite being one of the largest types of publicly available biomedical data. Here we review the potential of Generative models in recapitulating and extending biomedical knowledge from microarray data, which may thus limit the potential to yield hundreds of novel biomarkers. Here we review the potential of generative models and conduct a comparative analysis of VAE, GAN and gaussian mixture model (GMM) in a dataset focused on Tuberculosis. We further review whether previously known axes genes can be used as an effective strategy to employ domain knowledge while designing generative models as a means to further reduce biological noise and enhance signals that can be validated by standard enrichment approaches or functional experiments.

genomics↗

bioRxiv2020

lncRNA Mediated Hijacking of T-cell Hypoxia Response Pathway by Mycobacterium Tuberculosis Predicts Latent to Active Progression in Humans

Cytosolic functions of Long non-coding RNAs including mRNA translation masking and sponging are major regulators of biological pathways. Formation of T cell-bounded hypoxic granuloma is a host immune defence for containing infected Mtb-macrophages. Our study exploits the mechanistic pathway of Mtb-induced HIF1A silencing by the antisense lncRNA-HIF1A-AS2 in T cells. Computational analysis of in-vitro T-cell stimulation assays in progressors (n=119) versus non-progressor (n=221) tuberculosis patients revealed the role of lncRNA mediated disruption of hypoxia adaptation pathways in progressors. We found 291 upregulated and 227 downregulated DE lncRNAs that were correlated at mRNA level with HIF1A and HILPDA which are major players in hypoxia response. We also report novel lncRNA-AC010655 (AC010655.4 and AC010655.2) in cis with HILPDA, both of which contain binding sites for the BARX2 transcription factor, thus indicating a mechanistic role. Detailed comparison of infection with antigenic stimulation showed a non-random enrichment of lncRNAs in the cytoplasmic fraction of the cell in TB progressors. The lack of this pattern in non-progressors replicates indicates the hijacking of the lncRNA dynamics by Mtb. The in-vitro manifestation of this response in the absence of granuloma indicates pre-programmed host-pathogen interaction between T-cells and Mtb regulated through lncRNAs, thus tipping this balance towards progression or containment of Mtb. Finally, we trained multiple machine learning classifiers for reliable prediction of latent to the active progression of patients, yielding a model to guide aggressive treatment.

bioinformatics↗