Table I.

Continued.

Section	Input	Predictive Model	Output	Performance	Paper
3.4	Sequence Substrate structure Substrate physicochemical parameter	Gradient boost model regression	KM	MSE = 0.80 (log10-scale) R2 = 0.42	2021-Kroll (Kroll et al., 2021)
3.4	Sequence Substrate SMILES	CNN	kcat	Pearson r = 0.94 (log10-scale)	2022-Li (Li et al., 2022)
3.4	Sequence Substrate SMILES	Feed forward network	KD classifier	AUROC = 0.89	2022-Goldman (Goldman et al., 2022)
3.5	Sequence	Ridge regression	Fitness	MSE = 0.74	2020-Favor (Favor and Jayapurna, 2020)
3.5	Sequence	CNN Tweedie regression	Fitness	Spearman ρ = 0.61	2021-Wittmann (Wittmann et al., 2021b)
3.5	Sequence	Iterative MSA and conservation analysis	Conserved AA	N/A	2021-Teze (Teze et al., 2021)
3.5	Sequence	RNN	Fitness classifier Fitness regressor	AUROC = 0.88 Spearman ρ = 0.91	2021-Luo (Luo et al., 2021)
3.5	Sequence	Regularized linear regression	Fitness	Spearman ρ = 0.93	2021-Biswas (Biswas et al., 2021)
3.5	Sequence	Ridge regression	Fitness	Spearman ρ ~ 0.66	2022-Hsu (Hsu et al., 2022)
3.6	Sequence	Generative adversarial network	Artificial enzyme sequence	24% with catalytic activity	2021-Repecka (Repecka et al., 2021)
3.6	Sequence	Protein language model	Artificial enzyme sequence	AUC = 0.85	2021-Madani (Madani et al., 2021)
3.6	Sequence	Direct coupling statistical analysis of sequence MSA	Artificial enzyme sequence	Hit rate = 30%	2020-Russ (Russ et al., 2020)
3.6	Sequence	Variational autoencoder model of Blast sequence	Artificial enzyme sequence	Pearson R2 = 0.99	2022-Giessel (Giessel et al., 2022)

^aFor each research work, only the best-performing models are shown. Abbreviations: artificial neural network (ANN), convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN).

^bQ(von der Esch et al., 2019): Leave-one-out cross-fold validation score.