Modeling of Human Proteome by DMFold with DeepMSA2 MSA

This database lists structural models by DMFold for 5042 difficult proteins in human proteome. The model confidence is assessed by the predicted LDDT (pLDDT) score, where a higher pLDDT score signifies a higher model confidence. Usually, pLDDT score>0.7 indicates a correct fold. Here, two sets of DMFold models are listed: ‘v1’ refers to the models built with the 'pLDDT' pre-trained parameters and ‘v2’ to the models built with the 'pTM' pre-trained parameters. On average, the ‘v1’ models have a slightly higher pLDDT score than the ‘v2’ models, and users are suggested to select the models with a higher pLDDT score.