Improving chemical reaction yield prediction using pre-trained graph neural networks

Han, Jongmin; Kwon, Youngchun; Choi, Youn-Suk; Kang, Seokho

doi:10.1186/s13321-024-00818-z

Journal of Cheminformatics

Table 3 Comparison of predictive performance in terms of R\(^2\)

From: Improving chemical reaction yield prediction using pre-trained graph neural networks

Dataset	Split	Previous studies				Existing GNN pre-training methods					Proposed method
		MFF [4]	YieldBERT [6]	YieldBERT-DA [7]	YieldMPNN [8]	From-Scratch	MolCLR [13]	DGI [18]	ContextPred [21]	AttrMasking [21]	MolDescPred-MPNN	MolDescPred
Buchwald-Hartwig (Random Split)	70/30	0.932±0.008	0.951±0.005	0.969±0.004	0.974±0.001	0.971±0.002	0.974±0.001	0.974±0.001	0.974±0.001	0.974±0.002	0.974±0.002	0.974±0.001
	50/50	0.913±0.007	0.928±0.004	0.953±0.006	0.961±0.003	0.949±0.019	0.962±0.003	0.961±0.003	0.962±0.003	0.962±0.003	0.962±0.003	0.963±0.003
	30/70	0.878±0.010	0.882±0.011	0.917±0.010	0.934±0.008	0.923±0.010	0.937±0.007	0.934±0.008	0.935±0.008	0.935±0.008	0.936±0.008	0.937±0.008
	20/80	0.852±0.007	0.857±0.008	0.886±0.017	0.909±0.008	0.883±0.018	0.913±0.009	0.908±0.011	0.910±0.007	0.908±0.009	0.912±0.008	0.913±0.009
	10/90	0.791±0.011	0.793±0.016	0.818±0.009	0.841±0.013	0.763±0.032	0.842±0.016	0.839±0.017	0.837±0.014	0.839±0.020	0.838±0.014	0.847±0.016
	5/95	0.697±0.024	0.622±0.042	0.733±0.027	0.734±0.019	0.546±0.146	0.741±0.018	0.733±0.028	0.739±0.023	0.726±0.020	0.753±0.025	0.768±0.029
	2.5/97.5	0.576±0.047	0.436±0.034	0.604±0.031	0.628±0.062	0.391±0.194	0.636±0.051	0.616±0.061	0.583±0.082	0.623±0.042	0.619±0.042	0.662±0.053
	avg. rank	10.29±0.88	9.86±0.35	7.86±0.99	4.14±1.73	9.57±1.29	1.71±0.70	5.00±1.77	4.29±2.31	4.14±2.17	3.14±1.64	1.00±0.00
Suzuki-Miyaura (Random Split)	70/30	0.834±0.010	0.815±0.013	0.859±0.012	0.886±0.010	0.879±0.011	0.890±0.011	0.887±0.011	0.890±0.010	0.892±0.010	0.891±0.009	0.889±0.010
	50/50	0.810±0.006	0.780±0.009	0.823±0.007	0.867±0.003	0.855±0.004	0.869±0.004	0.867±0.005	0.870±0.004	0.869±0.004	0.870±0.004	0.869±0.004
	30/70	0.774±0.006	0.729±0.014	0.774±0.012	0.829±0.004	0.803±0.014	0.831±0.005	0.824±0.005	0.830±0.005	0.827±0.004	0.832±0.005	0.831±0.006
	20/80	0.738±0.013	0.676±0.015	0.719±0.022	0.794±0.011	0.735±0.035	0.794±0.010	0.783±0.012	0.790±0.012	0.788±0.011	0.797±0.010	0.794±0.007
	10/90	0.672±0.018	0.554±0.025	0.627±0.030	0.708±0.013	0.595±0.058	0.705±0.015	0.694±0.017	0.700±0.018	0.685±0.021	0.715±0.015	0.712±0.009
	5/95	0.592±0.022	0.430±0.040	0.491±0.034	0.565±0.018	0.454±0.103	0.542±0.048	0.573±0.020	0.566±0.021	0.520±0.038	0.601±0.021	0.594±0.016
	2.5/97.5	0.481±0.057	0.330±0.047	0.282±0.047	0.331±0.051	0.265±0.204	0.342±0.120	0.357±0.055	0.356±0.044	0.323±0.048	0.395±0.042	0.421±0.049
	avg. rank	7.00±3.30	10.57±1.05	9.29±0.45	5.14±1.81	9.14±1.12	3.86±1.81	5.71±1.16	4.00±1.41	5.71±2.60	1.43±0.73	2.57±1.05
Buchwald-Hartwig (Out-Of-Sample Split)	Test 1	0.882±0.004	0.824±0.010	0.811±0.047	0.744±0.042	0.609±0.086	0.876±0.023	0.755±0.023	0.756±0.051	0.859±0.018	0.827±0.011	0.883±0.009
	Test 2	0.727±0.006	0.829±0.037	0.866±0.020	0.876±0.026	0.877±0.021	0.882±0.026	0.816±0.056	0.877±0.030	0.892±0.017	0.866±0.038	0.913±0.010
	Test 3	0.650±0.006	0.741±0.030	0.585±0.067	0.717±0.024	0.610±0.081	0.603±0.034	0.631±0.019	0.658±0.049	0.650±0.013	0.648±0.026	0.761±0.028
	Test 4	0.388±0.008	0.444±0.077	0.157±0.034	0.496±0.031	0.420±0.186	0.481±0.020	0.224±0.016	0.252±0.071	0.471±0.032	0.455±0.042	0.382±0.045
	avg.rank	6.25±3.27	5.50±2.50	9.00±2.00	5.00±3.39	7.50±2.69	4.50±3.20	9.25±0.83	6.25±2.28	3.50±1.12	5.75±1.30	2.75±3.03

The best and second-best cases are highlighted in bold and underlined font, respectively

Back to article page

ISSN: 1758-2946

Contact us

Submission enquiries: journalsubmissions@springernature.com