A Deep Learning-Based Integrative Framework for Cancer Subtype Classification Leveraging Multi-Omics Data: A Software Engineering Approach

Abstract

The cancer subtypes classification is a significant aspect of oncology; it helps to predict the disease and effective treatment. High-throughput analysis techniques have been enhanced to enable researchers to generate large-scale multi-omics assemblies. Despite these enhancements, integrating heterogeneous molecular layers while maintaining interpretability is still challenging. This research attempt aims to present a deep learning framework for cancer subtype classification. The framework uses data from the Cancer Genome Atlas project, including genomic, epigenomic, and proteomic data. To develop the framework, a multimodal neural network architecture was used, where each omics pattern is processed via a dedicated branch before features are integrated. Furthermore, an advanced attention technique has been utilised to improve interpretability. The proposed framework reached an overall classification accuracy of 92.3%. For testing, reserved experimental data for breast and lung adenocarcinoma subtypes were set aside, and methods that perform better than common baseline approaches were used. In follow-up tests, the model stayed stable even when the data were noisy, which suggests that the same framework could be applied to other cancer subtypes, including colorectal cancer. These results suggest that multi-layer, interpretable deep learning models can support accurate cancer classification and more precise oncology that is grounded in biological evidence.