Stacked Deep Learning Ensemble for Multiomics Cancer Type Classification: Development and Validation Study

Area of Science:

• Oncology
• Bioinformatics
• Artificial Intelligence

Background:

• Cancer poses a significant global health burden, necessitating early and accurate diagnosis for effective treatment.
• This research focuses on classifying five common cancer types in Saudi Arabia: breast, colorectal, thyroid, non-Hodgkin lymphoma, and corpus uteri.

Purpose of the Study:

• To evaluate if integrating RNA sequencing, somatic mutation, and DNA methylation profiles in a stacking deep learning ensemble improves cancer classification accuracy.
• To compare the multiomics ensemble model's performance against state-of-the-art multiomics models and individual data types.

Main Methods:

• A stacking ensemble learning approach was employed, integrating support vector machine, k-nearest neighbors, artificial neural network, convolutional neural network, and random forest.
• The methodology included two main stages: data preprocessing (normalization, feature extraction) and ensemble stacking classification.

Main Results:

• The stacking ensemble model achieved 98% accuracy using multiomics data, outperforming models using individual data types (96% for RNA sequencing and methylation, 81% for somatic mutation).
• The findings suggest the viability of using multiomics data for cancer diagnosis in primary care settings.

Conclusions:

• Advanced machine learning techniques, specifically ensemble learning integrating multiomics data, significantly improve cancer detection and prognosis.
• This study provides valuable insights into the effectiveness of combining diverse biological data for more accurate cancer classification.