Skip to main content
. 2020 Nov 6;25(21):5172. doi: 10.3390/molecules25215172

Figure 2.

Figure 2

Two PCA components of the hidden features extracted by the GraphConvModel for training, test and prediction subsets. Green and blue marks correspond to the molecules used in training and test datasets for QSAR model construction. Purple marks correspond with all molecules used in virtual screening (DrugBank database) and yellow marks indicates the location of the best candidates obtained after molecular dynamic simulation: levothyroxine, amobarbital and ABP-700.