Fig. 2
From: Tpgen: a language model for stable protein design with a specific topology structure
Evaluation results for HHH generation. A The proportion of HHH structures in the generated sequences of each model (5000 sequences per model) was assessed using DSSP. TopoProGenerator generated the highest proportion of generated HHH structures,, significantly higher than that for the random baseline and RITA.. B The HHH structures (predicted by Omegafold) generated by each model were evaluated using the Rosetta energy scoring function. TopoProGenerator achieved the lowest score, indicating the highest level of stability. The X-axis represents the total energy score. C and D By utilizing Blast+, we compared each generated sequence with all three-helix datasets for identity matching, retaining the maximum identity value for each sequence (max_id_sequence). By employing the TM-score, we compared each predicted structure of the generated sequences with all three-helix datasets for identity matching, retaining the maximum identity value for each generated structure (max_id_structure)