Demystifying the black box: A survey on explainable artificial intelligence (XAI) in bioinformatics

Area of Science:

• Bioinformatics
• Artificial Intelligence
• Machine Learning

Background:

• Black-box AI/ML models lack transparency, hindering user confidence.
• Explainable AI (XAI) and explainability techniques are emerging to address this.
• Bioinformatics increasingly utilizes AI/ML, necessitating interpretable models.

Purpose of the Study:

• To review existing XAI techniques in bioinformatics, focusing on omics and imaging data.
• To analyze the demand for XAI in bioinformatics.
• To identify current XAI approaches, their limitations, and user-specific needs.

Main Methods:

• Literature review of explainability techniques in bioinformatics.
• Focus on applications within omics and imaging data analysis.
• Analysis of user requirements and system developer needs.

Main Results:

• Significant demand for XAI in bioinformatics, driven by transparency needs.
• Identification of current XAI methods and their limitations in omics and imaging.
• Emphasis on tailoring XAI to specific bioinformatics applications and user groups.

Conclusions:

• XAI is essential for trustworthy AI in bioinformatics.
• Further development of XAI methods is needed for omics and imaging data.
• Guidelines are provided for developers to create effective XAI systems.