TABLE 3.
A comparison of the reversible and standard ResNet architectures on folding and design tasks on CASP 11. For the folding task, we report both FM and TBM proteins dRMSD (Å) scores. For the design task, the KL-divergence of the PSSM and sequence (one-hot labels) are reported, along with the sequence prediction accuracy.
| Model | Folding | Design | ||
|---|---|---|---|---|
| FM | TBM | PSSM | One-hot | |
| MimNet (ResNet) | 7.1 | 5.9 | 1.00 | 1.77 (19.9%) |
| MimNet (Reversible) | 6.6 | 5.6 | 0.93 | 1.73 (20.7%) |