A deep learning framework for accurate reaction prediction and its application on high-throughput experimentation data

Table 3 Comparison of model performance over four separate reaction groups of our in-house dataset. The values of R, MAE, RMSE refers to the mean and standard deviation across the folds

Group	Size	Methods	R²	MAE	RMSE
G1	317	GraphRXN-concat	0.653 ± 0.085	0.08 ± 0.01	0.11 ± 0.01
		GraphRXN-sum	0.453 ± 0.145	0.11 ± 0.01	0.14 ± 0.02
		Yield-BERT	0.712 ± 0.070	0.07 ± 0	0.10 ± 0.01
		DeepReac +	0.544 ± 0.128	0.09 ± 0.01	0.13 ± 0.02
G2	419	GraphRXN-concat	0.628 ± 0.048	0.05 ± 0	0.07 ± 0.01
		GraphRXN-sum	0.590 ± 0.034	0.06 ± 0	0.07 ± 0
		Yield-BERT	0.512 ± 0.046	0.06 ± 0	0.08 ± 0.01
		DeepReac +	0.523 ± 0.059	0.06 ± 0	0.08 ± 0
G3	401	GraphRXN-concat	0.800 ± 0.030	0.06 ± 0	0.08 ± 0
		GraphRXN-sum	0.773 ± 0.020	0.06 ± 0	0.08 ± 0
		Yield-BERT	0.783 ± 0.012	0.06 ± 0	0.08 ± 0
		DeepReac +	0.744 ± 0.032	0.07 ± 0.01	0.09 ± 0.01
G4	421	GraphRXN-concat	0.445 ± 0.088	0.08 ± 0.01	0.12 ± 0.01
		GraphRXN-sum	0.405 ± 0.091	0.09 ± 0.01	0.12 ± 0.01
		Yield-BERT	0.490 ± 0.055	0.08 ± 0.01	0.11 ± 0.01
		DeepReac +	0.208 ± 0.17	0.10 ± 0.01	0.14 ± 0.02

ISSN: 1758-2946