Methodology for Good Machine Learning with Multi-Omics Data

Area of Science:

• Oncology
• Radiology
• Genomics
• Data Science

Background:

• A 4-year collaboration between Novartis Pharmaceuticals Corporation and the U.S. Food and Drug Administration (FDA) began in 2020.
• The project focuses on novel data modalities and advanced analytics.
• The primary scientific question concerns identifying prognostic and predictive factors for HR+/HER- metastatic breast cancer.

Purpose of the Study:

• To provide tangible guidelines for multi-omics projects involving multidisciplinary teams across institutions.
• To share insights gained from a collaborative project utilizing artificial intelligence (AI) and machine learning (ML).
• To offer actionable guidance for implementing exploratory data science projects.

Main Methods:

• Exploration of novel radio-genomics-based prognostic and predictive factors.
• Application of advanced analytics, including artificial intelligence and machine learning.
• Structured approach to multi-omics projects through four key steps: plan, design, develop, and disseminate.

Main Results:

• Valuable insights have been generated to facilitate future scientific projects.
• Demonstrated the potential of integrating complex data modalities and advanced analytics.
• Provided practical strategies for effective communication and good data science practices in collaborative research.

Conclusions:

• The collaboration has yielded significant insights into radio-genomics for metastatic breast cancer.
• The developed guidelines offer a framework for successful multi-omics research endeavors.
• Effective communication and robust data science practices are crucial for inter-institutional scientific projects.