Hundreds of Millions of AI-Generated Protein Maps... "Functional Fingerprints" Identified in Seconds [Reading Science]

Seoul National University Team Develops "FoldDisco," an Ultra-Fast Protein Structure Search Tool

Expected to Accelerate Discovery of Unknown Protein Functions and Facilitate Drug Design

A technology has been developed that can identify key clues determining protein function within hundreds of millions of protein structure data predicted by artificial intelligence (AI) in just a few seconds. It is expected that this technology will be useful not only for elucidating protein functions, but also for drug discovery and artificial enzyme design.

On June 11, the National Research Foundation of Korea announced that the research team led by Professor Martin Steineger at Seoul National University has developed a search tool called "FoldDisco," which enables ultra-fast exploration of structural motifs in massive protein structure databases (DB).

Operating Principle of FoldDisco. FoldDisco is a search technology that indexes the three-dimensional structure information of proteins to perform ultra-fast similarity searches within large-scale databases and determine matches of structural motifs.
Photo by Hyunbin Kim, Postdoctoral Researcher at Seoul National University

The results of this research were published online on June 5 in Nature Biotechnology, a leading international journal in the life sciences.

Protein structural motifs are three-dimensional structural patterns that determine protein function, such as enzyme active sites or binding sites. Researchers refer to these as the "three-dimensional fingerprints" of proteins. By identifying these fingerprints, it is possible to infer the roles of proteins whose functions are not yet known.

However, due to recent advances in AI-based protein structure prediction technology, the amount of data has surged, making it a new challenge to find meaningful motifs within vast structural datasets. Existing search methods require large amounts of storage and computation time, which limits their ability to analyze large-scale databases.

Ultra-Fast Search for Protein "Three-Dimensional Fingerprints"

The research team developed a method that indexes the distances, angles, and directional information between adjacent amino acid pairs in protein structures as numerical values. By also incorporating the directional information of amino acid side-chains, the system can distinguish even subtle structural differences in functional regions.

Additionally, by applying a "position-independent indexing" technique that does not store positional information and a "sparsity-based scoring" method that assigns higher weights to rare patterns, the team improved both search speed and storage efficiency.

As a result, FoldDisco was found to reduce index storage requirements to one-fourth of previous methods, while achieving a 20-fold increase in search speed and an 11-fold increase in indexing speed.

Research team photo. From left: Professor Martin Steiniger (corresponding author), Dr. Hyunbin Kim (first author), Researcher Seongeun Kim (co-author), Professor Milot Mirdita (co-author), Researcher Jaewon Yoon (co-author). Courtesy of Seoul National University

The research team also confirmed the effectiveness of their tool through real-world validation. They identified a "zinc finger" motif related to metal-binding function in proteins with previously unknown functions, and were able to accurately distinguish between active and inactive states of G protein-coupled receptors (GPCRs).

The researchers expect that this technology will be applied not only to elucidate unknown protein functions, but also broadly in the bio and pharmaceutical fields, including the design of artificial enzymes with specific active sites and the development of drug candidates.

Professor Steineger stated, "Currently, our focus is on protein structure searches, but in the future, we plan to expand the search scope to various biomolecules, such as nucleic acids and drugs, and develop this into a platform capable of integrative analysis of complex biological phenomena."

This content was produced with the assistance of AI translation services.