Fig. 2
From: High-throughput deep learning variant effect prediction with Sequence UNET
Base model prediction performance. A Pearson correlation between predicted and true PSSM values comparison PSSM prediction performance for Sequence UNET, a single layer CNN, SPBuild, ESM-1b logits, an ESM-1b top model and BLOSUM62. B ROC and PR curve AUC values comparing frequency classification performance of Sequence UNET with and without structural features, ESM-1b logits, an ESM-1b top model, a baseline single layer CNN, SIFT4G and BLOSUM62. All comparisons were made over the ProteinNet CASP12 test set, which contains 40 proteins with 12,094 total amino acids, giving 241,880 variant predictions with each tool