Compression of protein secondary structures enables ultra-fast and accurate structure searching
Lin, R.; Ahnert, S. E.
Show abstract
Protein structure prediction has undergone a revolution with the advent of AI-based algorithms, such as AlphaFold and RoseTTAFold. As a result, over 200 million predicted protein structures have been published. This wealth of structural data has created a need for rapid structure comparison algorithms, such as Foldseek, which enable efficient searches across this vast space of protein structures. Here we introduce a new ultra-compact representation of protein structure in the form of Secondary Structure Elements (SSEs). These are short sequences around 8% of the length and with 10% of the information content of full amino acid sequences and 3Di sequences. We show that, despite this compression factor, SSEs can be used as a highly effective tertiary structure comparison tool, with accuracy that approaches that of Foldseek, while offering a 200-fold speedup. In addition SSEs offer comparable performance to Foldseek in domain boundary retrieval. Furthermore we show that the particular way in which SSEs encode structure can also be used to specifically detect proteins that differ due to conformational change. These findings demonstrate that SSEs offer a valuable complementary approach for protein structure characterisation.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Scoring Protein Sequence Alignments Using Deep Learning 98%
- Embedding-based alignment: combining protein language models and alignment approaches to detect structural similarities in the twilight-zone 97%
- CATHe: Detection of remote homologues for CATH superfamilies using embeddings from protein language models 97%
Similar papers in this journal
Similar papers in this journal
- SAINT-Angle: self-attention augmented inception-inside-inception network and transfer learning improve protein backbone torsion angle prediction 97%
- Estimating Protein Complex Model Accuracy Using Graph Transformers and Pairwise Similarity Graphs 97%
- MULAN: Multimodal Protein Language Model for Sequence and Structure Encoding 96%
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.