MULTI-OMICS APPROACHES TO BREAST CANCER AND DEVELOPMENT OF GENERAL-PURPOSE TENSOR METHODS IN R

ENDOCRINE THERAPY RESISTANCE IN ERΑ-POSITIVE BREAST CANCER IS A COMPLEX REGULATORY PHENOTYPE THAT RARELY ARISES FROM ALTERATIONS CONFINED TO A SINGLE MOLECULAR LAYER. INSTEAD, RESISTANCE EMERGES THROUGH COORDINATED CHANGES ACROSS TRANSCRIPTIONAL PROGRAMS, CHROMATIN STATE, AND EPIGENETIC REGULATION, MOTIVATING INTEGRATIVE MULTI-OMICS APPROACHES THAT CONSTRAIN INTERPRETATION BY REQUIRING CONCORDANT EVIDENCE ACROSS COMPLEMENTARY ASSAYS. THIS THESIS PURSUES TWO OBJECTIVES: (I) TO INVESTIGATE ERΑ-ASSOCIATED REGULATORY DEPENDENCIES IN LUMINAL BREAST CANCER USING INTEGRATIVE FUNCTIONAL GENOMICS, INCLUDING THE DEVELOPMENT OF A PATHWAY-FIRST FRAMEWORK FOR CROSS-MODEL METHYLATION–EXPRESSION CONCORDANCE AND (II) TO CONTRIBUTE GENERAL-PURPOSE METHODOLOGY FOR SUPERVISED ANALYSIS AND INTERPRETATION OF LONGITUDINAL MULTI-OMICS DATA WITH AN INTRINSIC MULTI-WAY (TENSOR) STRUCTURE. IN PART I, BRPF1 IS INVESTIGATED AS A CHROMATIN-ASSOCIATED COREGULATOR CONNECTED TO ERΑ SIGNALLING. BY INTEGRATING ERΑ AND BRPF1 CHIP-SEQ, BRPF1 CHIP-SEQ AFTER ERΑ KNOCKDOWN, ATAC-SEQ UNDER BRPF1 PERTURBATION, AND RNA-SEQ, THE ANALYSES SUPPORT AN ERΑ-DEPENDENT RECRUITMENT OF BRPF1 AT A SUBSET OF SHARED REGULATORY LOCI AND SHOW THAT BRPF1 INHIBITION IS ASSOCIATED WITH WIDESPREAD ACCESSIBILITY LOSSES ENRICHED IN ERΑ-LINKED REGIONS, ACCOMPANIED BY COHERENT TRANSCRIPTIONAL ATTENUATION. TOGETHER, THESE MULTI-LAYER READOUTS MOTIVATE BRPF1 AS A CANDIDATE REGULATORY DEPENDENCY WITHIN ERΑ-ASSOCIATED CIRCUITRY. IN PART II, ENDOCRINE RESISTANCE ACROSS HETEROGENEOUS ERΑ-POSITIVE CELL LINE MODELS IS INVESTIGATED BY INTEGRATING EPIC DNA METHYLATION PROFILING WITH RNA-SEQ. BECAUSE GLOBAL METHYLATION STRUCTURE IS DOMINATED BY CELL LINE IDENTITY, THE ANALYSIS SHIFTS FROM CPG-LEVEL HETEROGENEITY TO PATHWAY-LEVEL CONVERGENCE, COMBINING HALLMARK GSEA, CONSENSUS LEADING-EDGE GENE CORES, AND PROMOTER METHYLATION SUMMARIES TO TEST WHETHER RECURRENT FUNCTIONAL PROGRAMS SHOW ENRICHED EPIGENETIC SUPPORT IN THE EXPECTED INVERSE DIRECTION. DIRECTION-PRESERVING FILTERS FURTHER ASSESS ROBUSTNESS BY ENFORCING INCREASING CROSS-MODEL CONSISTENCY AT BOTH THE PATHWAY AND GENE-CORE LEVELS, YIELDING CONSERVATIVE BUT INTERPRETABLE SETS OF RECURRENT PATHWAY SIGNATURES AND THEIR ASSOCIATED CONSENSUS GENE CORES. IN PART III, TENSORPLS IS INTRODUCED AS AN R PACKAGE FOR SUPERVISED ANALYSIS OF THREE-WAY LONGITUDINAL OMICS DATASETS REPRESENTED AS TENSORS. THE WORKFLOW COMBINES TENSOR-NATIVE PREPROCESSING AND TUCKER-3 IMPUTATION WITH PLS-BASED DISCRIMINATION, AND PROVIDES MODE-RESOLVED INTERPRETABILITY, INCLUDING TIME POINT CONTRIBUTION ANALYSIS AND TIME-RESOLVED FEATURE IMPORTANCE THROUGH VIP2D, WITH PERMUTATION-BASED ROBUSTNESS ASSESSMENT. EVALUATED ON LONGITUDINAL CASE–CONTROL DATA FROM THE TEDDY COHORT ACROSS MULTIPLE OMICS MODALITIES, TENSORPLS ILLUSTRATES HOW PRESERVING TENSOR STRUCTURE IN PREPROCESSING AND INTERPRETATION ENABLES STRUCTURED FEATURE PRIORITISATION AND IDENTIFICATION OF TEMPORAL WINDOWS ASSOCIATED WITH CLASS SEPARATION. TAKEN TOGETHER, THE CHAPTERS DEMONSTRATE THAT COMBINING CROSS-LAYER EVIDENCE WITH STRUCTURE-AWARE ANALYTICAL WORKFLOWS IMPROVES INTERPRETABILITY AND SUPPORTS ROBUST PRIORITISATION OF CANDIDATE MECHANISMS AND BIOMARKERS FOR SUBSEQUENT VALIDATION.