cancer_subtyping

cancer_subtyping stratifies cancer patients by molecular subtype using stacked denoising autoencoders (SdA) applied to integrated genomics, proteomics, and transcriptomics data to inform prognosis and biomarker discovery in clear cell renal cell carcinoma (ccRCC).

Key Features:

• Integration of Multi-Platform Genomic Data: Synthesizes genomics, proteomics, and transcriptomics datasets to capture the molecular heterogeneity of ccRCC.
• Unsupervised Deep Learning Approach: Employs stacked denoising autoencoders (SdA) to learn latent representations from unlabeled multi-omic data.
• Clinical Relevance: Distinguishes subtypes that differ in survival probability, histological grade, and pathological stage.
• Identification of Biomarkers: Identifies differentially expressed genes, proteins, and miRNAs between discovered subtypes.
• Transferability Across Cancer Types: Applies a model trained on ccRCC data to independent datasets such as Lung Adenocarcinoma (LUAD) and Low Grade Glioma (LGG).

Scientific Applications:

• Personalized Medicine: Stratifies patients into molecular subtypes to guide individualized diagnosis and treatment decisions.
• Prognostic Assessment: Provides subgroup-specific survival probability differences to inform prognosis and follow-up strategies.
• Biomarker Discovery: Enables discovery of subtype-associated genes, proteins, and miRNAs as candidate diagnostic or therapeutic targets.

Methodology:

Train stacked denoising autoencoders (SdA) on integrated genomics, proteomics, and transcriptomics datasets from ccRCC patients to reconstruct inputs while minimizing noise and extract latent features that define subtypes; correlate reconstructed inputs with original data and apply the trained model to independent LUAD and LGG datasets.