DNA Methylation-Based Cancer Classification

Deep neural networks read the DNA methylation profile of a tissue sample (Illumina 450k / EPIC beta values) and report the tissue of origin among 12 TCGA cancer types, a calibrated tumour probability with a three-zone call (normal, indeterminate, tumour) and an out-of-distribution warning. The genome-wide methylation profile is treated as a signal: 1-Mb windows are wavelet-denoised and turned into 118 signal features. The CpGs behind each call are shown with integrated gradients. A matching MATLAB version covers the signal processing and classification.

Category: Bioinformatics - Genomics and Sequence Analysis (Deep Neural Networks, Signal Processing, Explainable AI, MATLAB, Python)

Bioinformatics Project B06 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.