A comprehensive computational benchmark for evaluating deep learning-based protein function prediction approaches.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Accurate protein function annotation is vital for understanding life processes and disease mechanisms.
• Experimental methods cannot keep pace with the increasing number of known proteins.
• High-throughput sequencing generates vast biological data, enabling computational function prediction.

Purpose of the Study:

• To develop a comprehensive benchmark (BeProf) for evaluating computational protein function prediction methods.
• To assess the suitability of various algorithms across diverse application scenarios.
• To provide data-driven recommendations for selecting appropriate methods.

Main Methods:

• Collected and analyzed the latest protein datasets.
• Investigated and summarized 17 state-of-the-art computational methods.
• Developed a novel comprehensive evaluation metric and designed eight application scenarios.

Main Results:

• Evaluated the performance of existing methods across the designed scenarios.
• Identified strengths and weaknesses of different computational approaches.
• Provided a comparative analysis of protein function prediction algorithms.

Conclusions:

• BeProf serves as a valuable resource for benchmarking and selecting protein function prediction tools.
• The study offers practical guidance for researchers based on specific application needs.
• Facilitates more accurate and efficient protein function annotation in bioinformatics.