Skip to main content

Table 3 Comparison of the scoring powers of BsN-Score, BgN-Score, SNN-Score, Random Forests (RF), and the four top performing conventional SFs on four protein-family-specific tests sets.

From: BgN-Score and BsN-Score: Bagging and boosting based ensemble neural networks scoring functions for accurate binding affinity prediction of protein-ligand complexes

HIV protease (N= 112) Trypsin (N= 73)
Scoring function R p 1 R s 2 SD 3 RMSE4 D5 Scoring function R p 1 R s 2 SD 3 RMSE4 D5
X-Score::HPScore 0.341 0.339 1.54 1.509 N SYBYL::ChemScore 0.829 0.773 0.95 - U
BsN::XARG 0.290 0.230 1.56 1.705 Y DS::Ludi2 0.823 0.791 0.96 - U
RF::XARG 0.289 0.219 1.519 1.719 Y X-Score::HSScore 0.817 0.824 0.97 1.401 N
BgN-Score::XARG 0.287 0.209 1.58 1.860 Y DS::PLP2 0.797 0.774 1.02 - U
SYBYL::ChemScore 0.255 0.228 1.58 - U BgN-Score::XAR 0.776 0.719 1.06 1.070 Y
DrugScore::PairSurf 0.225 0.170 1.59 - U RF::XAR 0.774 0.753 1.07 1.133 Y
DS::PMF04 0.183 0.200 1.61 - U BsN-Score::AR 0.766 0.709 1.08 1.119 Y
SNN-Score::X 0.039 0.048 1.64 2.255 Y SNN-Score::X 0.735 0.672 1.14 1.209 Y
RF::XARG 0.964 0.975 0.44 0.588 N BsN-Score::XARG 0.937 0.920 0.59 0.678 N
BsN-Score::XARG 0.918 0.922 0.64 0.710 N RF::XARG 0.934 0.08 0.60 0.657 N
BgN-Score::XARG 0.848 0.808 1.02 1.024 N BgN-Score::XARG 0.892 0.848 0.76 0.805 N
SNN-Score::X 0.748 0.716 1.08 1.085 N SNN-Score::X 0.829 0.789 0.940 0.957 N
Carbonic anhydrase ( N = 44) Thrombin ( N = 38)
Scoring function R p 1 R s 2 SD 3 RMSE 4 D 5 Scoring function R p 1 R s 2 SD 3 RMSE 4 D 5
DS::PLP2 0.800 0.772 0.84 - U SNN-Score::X 0.756 0.704 1.38 1.433 Y
SYBYL::G-Score 0.706 0.646 0.99 - U BgN-Score::XARG 0.722 0.726 1.48 1.552 Y
SYBYL::ChemScore 0.699 0.631 1.00 - U BsN-Score::XARG 0.699 0.637 1.58 1.603 Y
BsN-Score::X 0.674 0.434 1.03 3.418 Y RF::XARG 0.697 0.693 1.52 1.674 Y
SNN-Score::X 0.631 0.451 1.08 3.561 Y DS::PLP1 0.667 0.672 1.58 - U
SYBYL::PMF-Score 0.627 0.618 1.09 - U SYBYL::G-Score 0.667 0.626 1.58 - U
BgN-Score::XA 0.625 0.423 1.09 3.642 Y X-Score::HSScore 0.666 0.586 1.58 1.737 N
RF::XARG 0.601 0.374 1.11 3.393 Y DrugScore::Pair 0.651 0.622 1.61 - U
BsN-Score::XARG 0.948 0.921 0.44 1.004 N BsN-Score::XARG 0.913 0.938 0.86 1.155 N
RF::XARG 0.910 0.860 0.57 1.140 N RF::XARG 0.910 0.934 0.86 1.125 N
BgN-Score::XARG 0.884 0.766 0.65 1.320 N BgN-Score::XARG 0.858 0.876 1.08 1.320 N
SNN-Score::X 0.652 0.310 1.05 1.687 N SNN-Score::X 0.761 0.756 1.37 1.374 N
  1. 1 R p is the Pearson correlation coefficient between predicted and measured BA values of complexes in this protein-family-specific test set.
  2. 2 R s is the Spearman correlation coefficient between predicted and measured BA values of complexes in this protein-family-specific test set.
  3. 3 SD is the standard deviation of errors between predicted and measured BA values of complexes in this protein-family-specific test set.
  4. 4 RMSE is the root-mean-square of errors between predicted and measured BA values of the test complexes in in this protein-family-specific test set. Test RMSE is not available for conventional SFs except for X-Score that we have re-constructed. Training RMSE is not reported in this table because the values are very similar to RMSEtrain in Table 1 due to the overlap between the training data sets of the two experiments.
  5. 5 This indicates whether the test set complexes are disjoint from (D = Y) or overlap with (D = N) the training set complexes for NN and RF models. Any overlap between the training and test data of the conventional SFs is unknown (D = U) to us.