Biology subjects

Natarajan, E.

Publications and source records attributed to Natarajan, E..

4 recordsLinked to original sources

bioRxiv2026

Deciphering the Tumor Microenvironment: An Integrated Single-Cell RNA-Seq and AI Framework for Novel Biomarker and Therapeutic Target Discovery in Melanoma

BackgroundMelanoma represents a highly immunogenic and therapeutically challenging malignancy. The complex cellular ecosystem of the tumor microenvironment (TME) acts as a critical driver of immune evasion, patient prognosis, and treatment resistance. Traditional bulk sequencing often fails to resolve these high-resolution intercellular dynamics. This investigation presents an end-to-end explainable machine learning and network biology framework designed to deconstruct TME single-cell heterogeneity, discover novel candidate biomarkers, and map actionable cell-cell communication networks. MethodologyHigh-quality single-cell RNA sequencing (scRNA-seq) expression data from melanoma lesions (GSE115978) were processed using a multi-phase computational workflow. Following cellular filtration, library size normalisation, and highly variable gene (HVG) selection, cells were partitioned using unsupervised Leiden clustering and annotated via signature gene scoring matrix methods (Wolf et al., 2018; Traag et al., 2019). An optimal gradient-boosted tree ensemble classifier (XGBoost) was constructed for cross-compartment biomarker screening (Chen & Guestrin, 2016), interpreted using SHAP (SHapley Additive exPlanations) values (Lundberg & Lee, 2017), and augmented with a PyTorch deep autoencoder. Downstream systems analyses included biological pathway enrichment via gseapy (Kuleshov et al., 2016), ligand-receptor communication mapping via LIANA (Efremova et al., 2020; Dimitrov et al., 2022), and Pearson co-expression network profiling. Cross-cohort validation was conducted on an independent melanoma cohort (GSE72056) (Tirosh et al., 2016b), and prognostic utility was clinically validated using empirical patient survival data from the TCGA-SKCM cohort (TCGA Research Network, 2015; Davidson-Pilon, 2019). ResultsUnsupervised Leiden clustering partitioned the cellular atlas into seven major structural and immunological compartments: Melanoma/Tumor, T-Cells, B-Cells, Macrophages, NK Cells, Endothelial cells, and Cancer-Associated Fibroblasts (CAFs). The trained XGBoost model prioritized CD79A, MLANA, LYZ, MFAP4, and CDH5 as top candidate biomarkers across distinct microenvironmental niches. SHAP explainability analysis confirmed MLANA and S100B as key positive predictors of tumor cell identity, while highlighting a bidirectional distribution for B2M linked to antigen presentation downregulation. Functional enrichment mapped Coagulation and Epithelial-Mesenchymal Transition (EMT) as highly dysregulated pathways driving the malignant state. Cell-cell communication profiling inferred highly significant SERPINE1 signalling axes to LRP1 and PLAUR receptors. SPARC was identified as the central co-expression hub regulator. Cross-cohort screening in GSE72056 confirmed biomarker stability across independent patient profiles. Empirical clinical survival validation within the TCGA-SKCM cohort (n=314) demonstrated that heightened expression of CD79A correlates with a statistically significant survival advantage (Log-Rank p = 4.4202 x 10-3), extending median overall survival from 26.7 months to 44.8 months. ConclusionThis computational pipeline systematically resolves TME heterogeneity, revealing a robust biomarker signature centred on CD79A and MLANA, alongside SERPINE1-driven immune-stromal crosstalk. The discovery of the protective prognostic role of CD79A links single-cell immune networks directly to clinical patient outcomes, providing a reproducible roadmap for anti-tumour immune engagement and immunotherapeutic stratification.

bioinformatics↗

bioRxiv2026

Graph Neural Networks (GNNs) for Protein-Ligand Interaction Prediction

Predicting protein-ligand interactions in the modern drug discovery has revolved from the involvement of artificial intelligence and structural bioinformatics using Graph Neural Networks (GNNs). The limited explainability of GNN models presents an important encumbrance in biomedical research, but it has achieved a high degree of accuracy in determining and identifying binding affinity and active compounds, as evidenced by [1] [2] [3] [4]. Here this research focuses on the interpretation of protein-ligand interactions at a molecular level, a rapidly developing area within Graph Neural Networks (GNNs). Now days modern study handling techniques such as visualization techniques, attention mechanism and model-based feature ascription by model to boost, and make robust and decrease false predictions on binding. Along with some approaches include like graph pooling strategies, message-passing optimization, self-supervised learning, transfer learning and contrastive learning are rapidly utilized to enhance the representative learnings. Furthermore, integration of molecular docking simulations, hybrid deep learning architectures and protein language model gives more reliable & biological predictions of protein-ligand interactions. That focuses on given process that identifies key ligand atoms and binding residues, as well as physicochemical factors influencing affinity, through chemical thought processes. Here this research work identified the challenges of developing biologically significant explanations, transparency, and the corollary dataset biases on interpretability. The research work conducted an in-depth investigation into the consolidation of protein language models to establish more reliable pathways for future research, examining hybrid architectures, transparent and energy-efficient GNNs, and scientifically grounded AI models for drug discovery. My research work highlights that XGNNs establishes a connection between Deep Learning and Biochemical expertise with increased confidence, which will enhance the accuracy of predictive models and computational models.

bioinformatics↗

bioRxiv2026

NETWORK-BASED FUNCTIONAL FRAGILITY REVEALS SYSTEM-LEVEL REORGANIZATION OF THE GUT MICROBIOME IN INFLAMMATORY BOWEL DISEASE

The human gut microbiome plays a critical role in host health, yet its functional organization in disease remains poorly understood. Most studies focus on taxonomic composition or pathway abundance, which fail to capture higher-order interactions governing system-level behavior. Here, we investigated microbiome functional organization in inflammatory bowel disease (IBD), including Crohns disease (CD), ulcerative colitis (UC), and healthy controls (HC), using a network-based framework across 60 metagenomic samples. Functional pathway profiles were used to construct correlation-based interaction networks, followed by analysis of network topology, functional redundancy, keystone pathway architecture, and system robustness. Disease-associated networks (CD and UC) exhibited reduced global connectivity, increased modular fragmentation, and centralization of keystone pathways, indicating a shift from distributed organization to more fragmented and fragile network structures compared to healthy controls. Notably, machine learning models demonstrated that network-derived features achieved higher classification performance (accuracy up to 0.824) compared to redundancy-based measures. These findings reveal that microbiome dysfunction in IBD is driven by large-scale reorganization of functional interaction networks rather than loss of functional capacity. This study highlights the importance of network-level analysis in understanding microbiome-associated disease and provides a systems-level framework for future research.

bioinformatics↗

bioRxiv2026

Identification and Analysis of Novel RNA Editing Sites in Neurodegenerative Diseases Using Machine Learning Approaches.

BackgroundRNA editing is a post-transcriptional modification that alters the sequence of an RNA transcript. Two types of RNA editing were found in mammals, involving the enzymatic deamination of either adenosine to inosine (A-to-I) or cytidine to uridine (C-to-U) nucleotides in RNA. A-to-I, which is the most common form of RNA editing, is mediated by the ADAR (adenosine deaminases acting on RNA) family of enzymes, ADAR1, ADAR2, and ADAR3. The editing event alters the hydrogen bond pairing of nucleobases, and the editing site will be recorded as guanosine rather than the original adenosine. Indeed, RNA editing deregulation has been linked to several nervous and neurodegenerative diseases. In this project work is done on Alzheimers disease (AD) and the samples are from anterior cingulate cortex of human brain tissue. AD is the main dementia in the world and a neurodegenerative condition prevalent in the elderly. MethodologyA total of 20 raw RNA-sequencing data samples containing 10 controls and 10 Alzheimers disease (AD) cases were collected from NCBI using SRA Toolkit. Quality assessment was performed using FastQC and processed using Trimmomatic. Alignment was done using STAR RNA-seq aligner. RNA editing detection was performed using REDItools, detected sites were subsequently annotated against the REDIportal database. The resulting control-specific and disease-specific novel editing sites were merged into a single dataset containing exclusively novel, group-specific A-to-I editing events. This merged dataset was subsequently used for downstream feature extraction and machine learning analysis. Probability-based filtering was done to extract high-confidence disease associated sites and their gene list was used for computational level biological validation, pathway and functional enrichment analysis as well as overlap with known AD loci. ResultsRandom Forest showed the highest accuracy score (0.804) and ROC-AUC score (0.854). Most important features that differentiated control and diseased novel sites in random forest were coverage ([~]0.35), editing level ([~]0.33) and GC content ([~]0.15). The AEI mean values is higher in both male and female diseased cases ([~]0.48-0.50) but less in male and female control cases ([~]0.14-0.21). The mean values of ADAR1_CPM higher in control cases (123.65-143.30) and is less in diseased cases (88.35-97.93), ADAR2_CPM is almost equal in all cases ([~]3.7-4.7) and ADAR3_CPM is very less in all the cases ([~]0-0.02). Most candidate editing site were present in exon ([~]62-67 %) CDS regions ([~]17-21%) and relatively smaller fraction of gene ([~]15-16 %). Editing alterations preferentially affect molecular systems governing synaptic structure, neurotransmission, and central nervous system integrity. In the main set -of the 2576 high-confidence genes identified, 33 overlapped with AD GWAS loci. In the core set -of the 1367 high-confidence genes identified, 11 overlapped with AD GWAS loci. ConclusionFeature like coverage, editing level and GC content contributed most. Alu sites are negligible as compared to non-alu sites but the AEI mean values are higher in diseased cases than in control cases. The mean values of ADAR1_CPM are higher than ADAR2_CPM and ADAR3_CPM.Sex does not play a major factor. High-confidence disease-associated RNA editing sites are strongly biased toward transcript-centric regions, particularly exons, with a notable subset affecting coding sequences. Importantly, enrichment of neurodegeneration-associated pathways and cognition-related human phenotypes further supports the disease relevance of these gene networks. RNA editing events in Alzheimers cortex may represent a regulatory mechanism largely independent of inherited genetic susceptibility loci.

neuroscience↗