. 2022 Oct 4;62(20):4852–4862. doi: 10.1021/acs.jcim.2c00715

Table 1. Unfiltered and Filtered Pre-Training Dataset Sizes and the Proportions They Represent of the Final Dataset^a.

dataset	GDB-13	Zinc 15	PubChem	ChEMBL	USPTO	total
unfiltered size (molecules)	977,468, 301	389,000,000	206,550,222	1,920,027	994,838	1,575,933,388
proportion of unfiltered total (%)	81	10	9	0.2	0.00825	100

The filtered aggregate dataset contains only unique SMILES, totaling 1,264,754,823 molecules.

Table 1. Unfiltered and Filtered Pre-Training Dataset Sizes and the Proportions They Represent of the Final Dataseta.