Interpretable and explainable predictive machine learning models for data-driven protein engineering.

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Protein engineering utilizes directed evolution and rational design to optimize protein properties.
• Artificial intelligence (AI) accelerates protein engineering via data-driven predictive models.
• Current AI models lack interpretability, limiting their real-world application and trustworthiness.

Purpose of the Study:

• To explore the principles and methodologies of Explainable Artificial Intelligence (XAI).
• To highlight XAI's relevance and potential in biotechnology and protein design.
• To propose theoretical pipelines integrating XAI with predictive models for protein engineering.

Main Methods:

• Review of XAI principles and methodologies.
• Analysis of XAI applications in biotechnology.
• Development of theoretical pipelines for integrating XAI in protein engineering.

Main Results:

• XAI offers insights into AI decision-making, enhancing model reliability.
• XAI application in protein engineering is underexplored but holds significant potential.
• Three theoretical XAI-integrated pipelines for protein design are proposed.

Conclusions:

• XAI can significantly enhance protein engineering by improving model interpretability and trustworthiness.
• Integrating XAI can guide machine learning-assisted directed evolution and protein design.
• Further research is needed to address challenges and develop XAI as a support tool for traditional protein engineering.