MOTLAB: A Weighted Multi-Omics Transfer Learning Approach to Mitigate Breast Cancer Racial Disparities.

Area of Science:

• Computational biology
• Bioinformatics
• Health disparities research

Background:

• Breast cancer (BC) disproportionately affects Black American women, leading to higher mortality rates compared to non-Hispanic White American women.
• Artificial intelligence (AI) and machine learning (ML), particularly transfer learning (TL), show promise in addressing BC health disparities by transferring knowledge from majority to minority groups.
• Existing TL models for BC often suffer from decreased performance due to limited clinical data and underutilization of multi-omics data integration.

Purpose of the Study:

• To develop and validate a novel weighted multi-modal transfer learning framework (MOTLAB) for optimizing multi-omics integration and mitigating racial disparities in breast cancer.
• To address limitations of existing methods, including insufficient patient samples and suboptimal integration of diverse omics data.
• To enhance the robustness and performance of AI-driven approaches for reducing breast cancer health disparities.

Main Methods:

• Developed MOTLAB, a weighted multi-modal transfer learning framework integrating mRNA, miRNA, and methylation data.
• Utilized patient-patient similarity (Pearson Correlation Coefficient) to construct a weighted integration of multi-omics data.
• Employed nested grid search for optimizing omics modality weights and Synthetic Minority Oversampling Technique (SMOTE) for data augmentation to address data imbalance.

Main Results:

• MOTLAB demonstrated superior performance in reducing breast cancer health disparities compared to existing multi-omics TL models.
• The framework outperformed single-omics, two-omics integration, conventional mixed, and independent models in mitigating BC health disparities.
• Optimized weighted integration of three omics data within MOTLAB significantly enhanced its effectiveness.

Conclusions:

• MOTLAB offers a robust and optimized approach for reducing racial disparities in breast cancer.
• The framework's ability to integrate and optimize multi-omics data holds significant potential for improving BC diagnosis, prognosis, and treatment.
• MOTLAB is extensible and can be applied to mitigate health disparities in other cancer types.