A unified approach to sequential and non-sequential structure alignment of proteins, RNAs, and DNAs

Many distantly related structure pairs exhibit structural similarities that can only be fully captured by a non-sequential alignment program. We present US-align2, a unified protocol for both sequential and non-sequential alignment of proteins and nucleic acids. On manually curated reference alignments for protein structural pairs with non-sequential relations, US-align2 achieves ≥13% higher agreement with reference alignments than existing sequential and non-sequential alignment methods. Non-sequential alignments also enabled US-align2 to have higher sensitivities in detecting RNA pairs from the same family with sequence identities <40%, obtaining ≥9% higher area under the receiver operating characteristic curve than third-party programs. The unique ability of US-align2 to parse both proteins and nucleic acids allows the method to detect protein-RNA and protein-DNA mimicries. Additionally, US-align2 performs full and semi-non-sequential alignments with at least 48% and 14% faster speed than existing programs for the same tasks, making it particularly useful for large-scale structural similarity detection.