FoldPAthreader is a new protein folding pathway prediction method that designs a new force field to guide the folding of amino acid chains into their native state through a series of transition states and potential intermediates. Given the query sequence of a target protein, three-dimensional structure is first modeled and the similar structures of the target are searched from the AlphaFold DB50 libraries. Then, structures similar to the target with TM-score ≥ 0.3 are selected for multiple structures alignment (MSTA), and residue frequency distribution information are extracted, which is further designed as a statistical potential energy function combined with the residue distance information extracted from the predicted structure. Meanwhile, a high-quality fragment library was constructed based on the candidate structures screened from MSTA. Finally, the protein folding pathway is simulated through three-stage conformational sampling based on fragment assembly guided by statistical and physical potential energy force fields using different energy terms and weights.

PAthreader is a method for the remote homologous template recognition. First, the structure profile is extracted from PAcluster80, a template library constructed by clustering PDB and AlphaFold DB with a threshold of 80% structural similarity. Meanwhile, the distance profile is predicted by our in-house DeepMDisPre. Then, a three-track alignment algorithm is proposed to align the query sequence to each cluster seed to get the maximum alignment score (alignScore). As an supplement to alignScore, the physical and geometric features of the alignment structure are extracted and fed into a convolutional network with self-attention to predict DMScore (pDMScore), a global structure scoring metric linearly weighted with the alignScore for the template ranking.