Table 5 PPB prediction results of synthetic cyclic peptides for sparse modeling by ELS and FBS compared to the baseline results. These situations are shown in a part of Fig. 1(b). The values with asterisk represent the best prediction performance in each evaluation criterion, and ridge-CP and lasso-CP lines are reproduced from Table 3
Method | Training set | Test set | RMSE (fb) | MAE (fb) | R (ln Ka) |
|---|---|---|---|---|---|
ridge-SM (baseline #1) | Small molecules (training data) | Synthetic cyclic peptides (# = 16) | 0.321 | 0.263 | 0.761 |
lasso-SM (baseline #2) | Small molecules (training data) | Synthetic cyclic peptides (# = 16) | 0.276 | 0.228 | 0.714 |
ELS | Small molecules (training data) | Synthetic cyclic peptides (# = 16) | 0.319 | 0.269 | 0.748 |
FBS | Small molecules (training data) | Synthetic cyclic peptides (# = 16) | 0.230* | 0.194* | 0.805* |
ridge-CP (baseline #3) | Cyclic peptide drugs (# = 24) | Synthetic cyclic peptides (# = 16) | 0.413 | 0.354 | 0.442 |
lasso-CP (baseline #4) | Cyclic peptide drugs (# = 24) | Synthetic cyclic peptides (# = 16) | 0.688 | 0.627 | 0.069 |