PSIST: indexing protein structures using suffix trees.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Protein structure analysis is crucial for predicting function and aiding drug discovery.
• Existing methods for searching protein structures can be computationally intensive and lack scalability.

Purpose of the Study:

• To develop a new, efficient method for extracting local feature vectors from protein structures.
• To implement a scalable search algorithm for identifying similar protein structures in large databases.

Main Methods:

• Representing each amino acid residue as a triangle, using distances between C-alpha atoms and inter-plane angles to define local features.
• Indexing normalized local feature vectors using suffix trees for rapid maximal match retrieval.
• Chaining maximal matches to align query structures with database proteins and ranking by alignment score.

Main Results:

• Achieved high classification accuracy: up to 97.8% at the superfamily level and 99.4% at the class level (SCOP classification).
• Demonstrated effective retrieval, with an average of 7.49 out of 10 proteins from the same superfamily found within the top 10 matches.
• The method's performance is competitive with existing state-of-the-art approaches.

Conclusions:

• The developed method provides an effective and scalable approach for protein structure comparison and classification.
• Suffix tree indexing of local feature vectors significantly enhances the speed and accuracy of searching for similar protein structures.
• This technique holds promise for advancing protein function prediction, classification, and drug discovery efforts.