Machine Learning Approaches for Quality Assessment of Protein Structures

Area of Science:

• Structural bioinformatics
• Computational biology
• Biomedical research

Background:

• Accurate protein structures are vital for drug discovery and design.
• Experimental structure determination is costly and time-consuming.
• Computational protein structure prediction methods require quality assessment.

Purpose of the Study:

• To systematically review major machine learning-based methods for estimating model accuracy (EMA) in protein structures.
• To discuss the significance of these methods from a methodological perspective.
• To provide an introductory guide to modern protein quality assessment research.

Main Methods:

• Review of machine learning-based EMA methods developed in the past ten years.
• Grouping methods by machine learning approach: support vector machine, artificial neural networks, ensemble learning, and Bayesian learning.
• Description of the background of EMA, including the CASP challenge and evaluation metrics, and major ML/DL techniques.

Main Results:

• Machine learning-based EMA methods have shown top performance in CASP challenges.
• Discussion of the methodological significance of various ML approaches for EMA.
• Identification of trends and advancements in ML-driven protein quality assessment.

Conclusions:

• Machine learning approaches are highly effective for estimating protein model accuracy.
• This review offers insights into current ML-based EMA methods and future research directions.
• Understanding EMA is critical for advancing computational drug discovery and design.