Some of these appear to be clear mistakes; there are three predictions of log S between 4.1 and 6.0 for molecules where the median predictions were between − 3.25 and − 4.5.
There are eight prediction units (PUs) shown in Fig. 1, and these eight PU partitions are named as PART_2N×2N, PART_N×N, PART_2N×N, PART_N×2N, PART_2N×nU, PART_2N×nD, PART_nL×2N, PART_nR×2N, respectively.
If these three best PhModels were selected, there are nine prediction results and then there are 2 − 1 = 511 combinations.
Therefore, there are nine prediction results (denoted as Ztrain × Ztest, Z = {Best, Fast, Caesar}, that is, BesttrainBesttest) generated for testing set inhibitors.
Therefore, there are six prediction results (Z1train × Z2test, Z1 = {Best, Fast} and Z2 = {Best, Fast, Caesar}) and 2 − 1 = 63 combinations.
Therefore, there are six prediction results, BesttrainBesttest (denoted as BB), BesttrainFasttest (denoted as BF), BesttrainCaseartest (denoted as BC), FasrtrainBesttest (denoted as FB), FasttrainFasttest (denoted as FF), and FasttrainCaseartest (denoted as FC), for testing set inhibitors.
Burnley will get used to being tipped for relegation, and I am not about to buck the trend - but if there is one prediction I wish to get hopelessly wrong it is this one.
The final test would require having a sextuple mutant (cca1 lhy rve8 rve6 rve5 rve4) and then there would be two predictions.
There are two main prediction ways: 1) causal prediction [27, 30] and 2) noncausal prediction [28, 29].
