GLMYsymm: inferring symmetry categories of protein complexes from single sequences using persistent GLMY homology

Proteins often perform essential biological functions in the form of complexes. The assembly of protein complexes typically exhibits symmetry, which contributes to a more stable structural organization. Therefore, investigating structural symmetry is of great importance for protein complex structure prediction. However, existing methods for predicting protein structural symmetry are limited in number and generally show unsatisfactory accuracy. To address this issue, we propose GLMYsymm, a single-sequence-based model for symmetry prediction of homologous protein complexes. The key innovation of this work lies in the use of persistent Grigor'yan-Lin-Muranov-Yau (GLMY) homology to extract topological information from protein sequences, which is further integrated with the Evolutionary Scale Modeling 2 (ESM2) pretrained model to enable end-to-end symmetry prediction directly from sequence data. To the best of our knowledge, this is the first study to apply persistent GLMY homology to protein sequence feature extraction, without relying on structural information or multiple sequence alignment. Through extensive ablation and comparative experiments, we further demonstrate the effectiveness of GLMY homology-based topological sequence features for training deep learning models. Furthermore, the GLMYsymm framework outperforms existing sequence-based methods, achieving an improvement of approximately 0.32 in Macro area under the precision-recall curve (AUC-PR) over Seq2Symm and QUEEN models on the same test dataset. In the field of protein structure prediction, GLMYsymm can be used to assess the symmetry category of predicted protein structures, thereby assisting in protein structure quality evaluation, and can also serve as an important reference for stoichiometry prediction. The code and datasets for GLMYsymm are available at http://mialab.ruc.edu.cn/GLMYsymmServer/.