Detection of circular permutations by Protein Language Models

Protein circular permutations are crucial for understanding protein evolution and functionality. Traditional detection methods face challenges: sequence-based approaches struggle with detecting distant homologs, while structure-based approaches are limited by the need for structure generation and often treat proteins as rigid bodies. Protein Language Model-based alignment tools have shown advantages in utilizing sequence information to overcome the challenges of detecting distant homologs without requiring structural input. However, many current Protein Language Model-based alignment methods, which rely on sequence alignment algorithms like the Smith-Waterman algorithm, face significant difficulties when dealing with circular permutation (CP) due to their dependency on linear sequence order. This sequence order dependency makes them unsuitable for accurately detecting CP. Our approach, named plmCP, combines classical genetic principles with modern alignment techniques leveraging Protein Language Models to address these limitations. By integrating genetic knowledge, the plmCP method avoids the sequence order dependency, allowing for effective detection of circular permutations and contributing significantly to protein research and engineering by embracing structural flexibility.