Obtaining genetics insights from deep learning via explainable artificial intelligence

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Deep learning models are state-of-the-art for functional predictions in genomics.
• The decision-making process of these AI models is often opaque.
• Understanding AI predictions is crucial for generating new biological insights.

Purpose of the Study:

• To review the emerging field of explainable AI (xAI).
• To highlight the potential of xAI in empowering life science researchers.
• To provide mechanistic insights into complex deep learning models.

Main Methods:

• Categorization of model interpretation approaches.
• Explanation of how each xAI approach works.
• Discussion of assumptions and limitations of xAI methods.

Main Results:

• xAI offers a pathway to understanding AI-driven genomics predictions.
• Various interpretation techniques exist, each with unique strengths and weaknesses.
• Contextualizing xAI within high-throughput biological data is essential.

Conclusions:

• Explainable AI is vital for unlocking the full potential of AI in genomics research.
• Mechanistic insights from xAI can drive novel discoveries in genetic processes.
• Researchers must carefully consider the assumptions and limitations of xAI methods.