Automated functional classification of experimental and predicted protein structures

Area of Science:

• Structural bioinformatics
• Computational biology
• Protein science

Background:

• Proteins with similar sequences or structures can exhibit diverse functions.
• Inferring protein function solely from sequence or structural similarity is often unreliable.

Purpose of the Study:

• To evaluate the limitations of homology-based function assignment for proteins in multi-functional families.
• To assess the performance of the Functional Signatures from Structural Alignments (FSSA) algorithm for protein function classification.

Main Methods:

• Analysis of experimental protein structures from the Structural Classification of Proteins (SCOP) database.
• Evaluation of predicted protein structures from LiveBench and PDB-CAFASP experiments.
• Application and performance comparison of the Functional Signatures from Structural Alignments (FSSA) algorithm.

Main Results:

• Approximately half of experimental structures belong to multi-functional fold families, limiting function inference from similarity alone.
• Homology-based functional assignments were inaccurate in about one-third of cases for proteins in multi-functional families.
• The FSSA algorithm demonstrated superior accuracy compared to homology-based methods for both experimental and predicted structures.

Conclusions:

• The FSSA algorithm offers improved accuracy for protein structure function classification, particularly for multi-functional families.
• FSSA's effectiveness stems from its use of local structural information rather than global similarity.
• The FSSA algorithm is available as a webserver for broader accessibility.