Machine Learning to Advance Human Genome-Wide Association Studies

Area of Science:

• Genetics
• Epidemiology
• Computer Science

Background:

• Machine learning (ML) methods, including deep learning, reinforcement learning, and generative artificial intelligence, offer powerful tools for analyzing complex biological systems.
• While ML has been applied to human genetic epidemiology since 2004, its full potential remains largely untapped.

Purpose of the Study:

• To review the main applications of ML in assigning human genetic loci to health outcomes.
• To summarize widely used ML methods, discussing their advantages and challenges in genetic research.
• To identify and evaluate tools designed for hypothesis-free analysis of genetic variation data.

Main Methods:

• Review of existing literature on machine learning applications in human genetic epidemiology.
• Summary and discussion of common ML techniques used for genetic data analysis.
• Identification and assessment of specialized tools like Combi, GenNet, and GMSTool.

Main Results:

• ML methods provide efficient ways to decipher information from large datasets for understanding complex biological systems.
• Several tools (Combi, GenNet, GMSTool) facilitate hypothesis-free analysis of genetic variation data, integrating various ML methods.
• The review discusses the added value and limitations of these tools from a geneticist's viewpoint.

Conclusions:

• Machine learning holds significant promise for advancing human genetic epidemiological research, particularly in linking genetic variations to health outcomes.
• The development and application of specialized tools are crucial for maximizing the utility of ML in genetic research.
• Future directions include the integration of foundation models and large multi-modal omics biobank initiatives for more comprehensive genetic analyses.