Journal of Cheminformatics

Table 4 Performance (r²) of pairing strategies

From: Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantification

Model	Lipophilicity	Freesolv	ESOL
MLP-FP	0.61	0.82	0.84
MLP-ΔFP	0.62^a (0.59)^b	0.82^a (0.81)^b	0.85^a (0.82)^b
MLP-SNN	0.64^a (0.64)^b	0.84^a (0.83)^b	0.86^a (0.83)^b
RF-FP	0.58	0.75	0.81
RF-ΔFP	0.62	0.81^a	0.84^a
Chemformer	0.76	0.91	0.92
Chemformer-SNN	0.74^a (0.62)^c	0.91^a (0.90)^c	0.86^a (0.80)^c

^a Trained with the similarity-based pairing
^b Trained with exhaustive pairing
^c Trained with the random paring where each compound was paired with 50 randomly chosen compounds as an approximation of exhaustive pairing

Back to article page

ISSN: 1758-2946

Contact us

Submission enquiries: journalsubmissions@springernature.com