Table 1 Prediction performance with different models (including previously reported GBRT²³, SMFluo²⁹, UVVisML⁴⁵, SchNet²⁶, and ABT-MPNN⁴⁶) towards FluoDB

Object	Algorithms	MAE	MSE	RMSE	R²
λ_abs	GBRT	13.67	824.24	28.71	0.93
	SMFluo	21.19	1255.71	35.44	0.89
	UVVisML	13.94	716.91	26.78	0.94
	SchNet	22.17	1684.74	41.05	0.63
	ABT-MPNN	12.66	687.97	26.23	0.94
	FLSF_MACCS	12.96	713.33	26.71	0.94
	FLSF_Morgan	14.75	853.52	29.22	0.92
	FLSF	12.56	675.34	25.99	0.94
λ_em	GBRT	14.56	671.52	25.91	0.92
	SMFluo	27.82	1467.36	38.31	0.83
	UVVisML	13.98	518.02	22.76	0.94
	SchNet	38.26	2695.06	51.91	0.43
	ABT-MPNN	13.30	521.65	22.84	0.94
	FLSF_MACCS	13.88	560.29	23.67	0.94
	FLSF_Morgan	15.66	746.21	27.32	0.92
	FLSF	13.27	545.12	23.35	0.94
Φ_PL	GBRT	0.12	0.03	0.18	0.68
	SMFluo	0.13	0.04	0.21	0.57
	UVVisML	0.13	0.04	0.19	0.64
	SchNet	0.15	0.04	0.20	0.39
	ABT-MPNN	0.12	0.03	0.19	0.65
	FLSF_MACCS	0.13	0.04	0.20	0.61
	FLSF_Morgan	0.12	0.04	0.19	0.64
	FLSF	0.12	0.03	0.19	0.66
ε_max	GBRT	0.20	0.10	0.31	0.66
	SMFluo	0.22	0.14	0.37	0.53
	UVVisML	0.26	0.14	0.37	0.51
	SchNet	0.51	0.51	0.71	-2.01
	ABT-MPNN	0.32	0.20	0.45	0.31
	FLSF_MACCS	0.25	0.13	0.36	0.56
	FLSF_Morgan	0.23	0.11	0.33	0.61
	FLSF	0.23	0.12	0.34	0.59

The unit of MAE/RMSE is nm for λ_abs and λ_em, not applicable for Φ_PL and ε_max (used in log₁₀ε_max). GBRT, SMFluo, UVVisML, SchNet, and ABT-MPNN are implemented from their open-source codes. The difference between FLSF (FLuorescence prediction with fluoroScaFfold-driven model) and FLSF_MACCS/FLSF_Morgan is the feature extraction method, as fluoroscaffold was introduced in FLSF, while the conventional MACCS/Morgan fingerprints were used for the latter.
λ_abs maximum absorption wavelength, λ_em maximum emission wavelength, Φ_PL photoluminescence quantum yield, ε_max molar absorption coefficient, MAE mean absolute error, MSE mean-square error, RMSE root-mean-square error, R² the coefficient of determination.

Quick links

Search