LOGICS: Learning optimal generative distribution for designing de novo chemical structures

Table 1 Statistics and descriptions of the compound data used in the generator pre-training and the predictor construction

Datasets	Total	Fivefold CV^a	Test	Active threshold
Pre-training data (Generator)	1,583,425	−	−	−
Training & testing data (Predictor)
KOR	3881	3230	651	> 7.0 pIC₅₀
PIK3CA	1462	1215	247	> 8.0 pKx

ISSN: 1758-2946