INTELLIGENT MULTI OMICS INTEGRATION FOR BREAST CANCER MOLECULAR SUBTYPE CLASSIFICATION

Abstract

Breast cancer is a heterogeneous disease with distinct molecular subtypes, making accurate subtype classification an important challenge in computational oncology. Conventional single-omics approaches often provide limited biological information by relying on individual molecular data sources. This study proposes MO-BCNet, a hybrid CNN-BiLSTM multi-omics deep learning framework for breast cancer molecular subtype classification. The proposed approach integrates complementary mRNA and miRNA expression profiles to learn informative molecular representations through convolutional feature extraction and bidirectional sequential learning. Experiments were conducted on a publicly available BRCA multi-omics dataset containing 671 samples and 5,366 integrated features, which were reduced to 5,200 informative features during preprocessing. The proposed model achieved an accuracy of 95.75%, precision of 95.8%, recall of 95.8%, F1-score of 95.7%, and AUC-ROC of 99.5%. The results demonstrate strong discriminative performance, stable convergence, and effective classification of breast cancer molecular subtypes. Overall, the findings indicate that integrating complementary multi-omics information with hybrid deep learning can provide an effective computational framework for breast cancer classification and support future research in biomarker discovery and precision medicine.

Keywords:  Breast Cancer, Multi-Omics, Molecular Subtype Classification, mRNA, miRNA, CNN, BiLSTM, Deep Learning, Precision Medicine