Seoul National University team develops ultrafast tool to search 53 million protein structures in seconds

A search tool capable of analyzing the key structures of tens of millions of proteins in just a few seconds has been developed.

The National Research Foundation of Korea announced Thursday that a research team led by Martin Steinegger, a professor at Seoul National University, has developed Folddisco, a tool that rapidly scans protein structures to identify structural motifs.

Since the emergence of AlphaFold2, AI has been generating hundreds of millions of protein structures in a matter of days — work that once took decades of laboratory experiments — accelerating advances in protein design, artificial enzymes and new drug development.

A protein's structural motif is a three-dimensional pattern, such as an enzyme's active site or binding site, that is short in length but plays a decisive role in function.

Formed by a handful of amino acids arranged at precise positions and angles, these structural motifs act as a kind of "fingerprint" for identifying protein function. Finding that fingerprint can help researchers infer the role of an otherwise unknown protein. Existing structural indexing methods, however, required enormous storage space and time, making them impractical for large-scale databases.

To overcome this limitation, the research team developed an approach that quantifies and indexes the distance, angle and orientation of adjacent amino acid pairs within a three-dimensional structure.

By adding side-chain orientation data to the geometric features of amino acid pairs, the tool can precisely distinguish even subtle shape variations at functional sites.

The team also combined a "position-free indexing" technique — which stores no positional data — with a "sparsity-based" scoring method that assigns greater weight to rare patterns, boosting both speed and storage efficiency simultaneously.

As a result, Folddisco detects structural motifs 20 times faster in searches and 11 times faster in index generation than existing methods, while using only one-quarter of the index storage space.

Using Folddisco, the research team identified zinc finger motifs in proteins whose functions had previously been unknown and clearly distinguished between the active and inactive states of G protein-coupled receptors (GPCRs).

Folddisco can search everything from very short active-site motifs to structurally distant patterns. The tool is expected to find broad application across the biotech and pharmaceutical fields — from research into the functions of unknown proteins to the design of artificial enzymes with specific active sites and the development of new drug candidates.

"Folddisco can be used in a wide range of applications, such as inferring the function of a protein that is difficult to identify by sequence alone, or distinguishing the state of a receptor that is a drug target," Steinegger said. "In particular, when designing artificial enzymes or new drugs, it can search vast structural databases for patterns resembling a specific active site, making it a search engine for protein design research."

Steinegger said the tool is currently limited to protein structure searches and cannot yet handle other biomolecules — such as nucleic acids or small-molecule drugs — that interact with proteins. "We plan to expand the search scope to biomolecules broadly and develop it into an integrated tool capable of analyzing complex biological phenomena from multiple angles," he said.

The research, supported by the Ministry of Science and ICT and the National Research Foundation of Korea, was published in the international life sciences journal Nature Biotechnology.