Research Projects

Rationale: Breast cancer is a leading cause of cancer-related mortality among women worldwide and is marked by pronounced molecular heterogeneity that impacts prognosis and therapeutic outcomes. Accurate classification into intrinsic subtypes: Luminal A, Luminal B, HER2-enriched, and Basal is essential for guiding personalized treatment strategies. While mRNA-based signatures are widely adopted, emerging evidence highlights the promise of long non-coding RNAs (lncRNAs) as underexplored yet highly informative molecular markers. lncRNAs regulate diverse processes including transcription, chromatin remodeling, and RNA–protein interactions, and exhibit subtype- and tissue-specific expression patterns, making them ideal candidates for refined molecular stratification. This proposal aims to systematically harness both expression and sequence-level features of lncRNAs for breast cancer subtype classification using advanced artificial intelligence (AI) approaches. We will employ traditional machine learning alongside deep learning architectures, with a particular emphasis on transformer-based Large Language Models (LLMs), to construct predictive, interpretable, and biologically grounded classification frameworks. LLMs will be utilized to derive contextual embeddings from raw lncRNA sequences, enabling discovery of latent regulatory ‘grammar’ elements that differentiate cancer subtypes. Central hypothesis: Specific lncRNAs encode subtype-distinctive regulatory signatures that can be uncovered through advanced computational modeling, particularly by using Large Language Models (LLMs) to extract contextual patterns from nucleotide sequences and further interpreted through integrative analysis of genomic, epigenomic, and structural features. Scientific Objectives: 1. To identify lncRNAs differentially expressed across subtypes using TCGA-BRCA transcriptomic data. 2. To develop and benchmark multiple machine learning classifiers (Random Forest, XGBoost, Neural Networks) for subtype prediction. 3. To implement LLM-based models to classify subtypes using lncRNA expression and sequence inputs. 4. To prioritize robust and interpretable lncRNA biomarkers using SHAP-based feature selection. 5. To map the genomic and epigenomic contexts of these lncRNAs, including enhancer overlaps, DNA methylation patterns, and motif enrichments. 6. To construct lncRNA–transcription factor–gene co-expression networks and propose regulatory circuits. 7. To experimentally validate selected high-confidence lncRNAs using qRT-PCR in representative breast cancer cell lines. Expected significance: This study will deliver a concise panel of subtype-specific lncRNA biomarkers with diagnostic and therapeutic relevance. By integrating LLM-based sequence modeling, expression-driven classification, and functional validation, it advances the frontier of AI-powered precision oncology.