. 2021 Nov 3;6(45):30743–30751. doi: 10.1021/acsomega.1c04778

Table 1. Accuracy Scores (%) for Classification of Reactions with an External LightGBM Classifier on Top of Projections Based on Difference Fingerprints From the RDKit and BERT FP^a.

	fingerprint and descriptors types
perplexity	MorganFP	AtomPairFP	Topological Torsion	BERT FP
10	84.0	83.1	86.4	78.1
30	83.5	82.8	85.8	77.2
100	82.9	81.3	85.1	76.0
500	79.2	75.5	81.4	71.8
multi-scale	84.3	83.1	87.0	77.5

A value in bold is the best score. The accuracy scores correspond to models trained for 80 epochs.

Table 1. Accuracy Scores (%) for Classification of Reactions with an External LightGBM Classifier on Top of Projections Based on Difference Fingerprints From the RDKit and BERT FPa.