FoldExplorer: Fast and Accurate Protein Structure Search with Sequence-Enhanced Graph Embedding

The advent of highly accurate protein structure prediction methods has fueled an exponential expansion of the protein structure database. Consequently, there is a rising demand for rapid and precise protein structure search. Traditional alignment-based methods are designed for precise pairwise comparisons, offering high accuracy. However, they face challenges when searching within large databases. In response to this challenge, we propose a novel deep-learning approach FoldExplorer. It leverages graph attention neural networks and protein language models to jointly encode structural and sequence information, generating embeddings tailored for protein structure search. FoldExplorer achieves competitive performance in geometric similarity search and classification tasks, outperforming recent deep learning and sequence-based methods, and approaching classical alignment tools. Significantly, FoldExplorer remains effective when searching low-confidence predicted structures. Meanwhile, FoldExplorer is particularly highly efficient when searching in large-scale databases. The accurate embedding space generated by FoldExplorer enables providing a comprehensive protein structure space view, which will provide novel cluster and boundary insights on the protein universe studies. A publicly accessible search web server is available at: http://www.csbio.sjtu.edu.cn/bioinf/FoldExplorer/.