Figure 1
Prediction accuracy compared across models. (a) Average accuracy per protein for each model. The average accuracy is 60.74%, 64.4%, 64.8%, and 68.3%, respectively, for the ESM1b, CNN, RESNET, and BERT models. Average accuracies are highlighted by the black point within each violin. (b) Correlation between CNN and ESM1b accuracy (\(r = 0.09\), \(p = 0.28\)). (c) Correlation between CNN and BERT accuracy (\(r = 0.22\), \(p = 0.008\)). (d) Correlation between CNN and RESNET accuracy (\(r = 0.77\), \(p < 10^{-10}\)). (e) Correlation between RESNET and ESM1b accuracy (\(r = 0.12\), \(p = 0.16\)). (f) Correlation between RESNET and BERT accuracy (\(r = 0.20\), \(p = 0.016\)). (g) Correlation between ESM1b and BERT accuracy (\(r = 0.52\), \(p = 10^{-10}\)). Each point represents a single protein.
