Fig. 4: Model development and validation. | Nature Communications

Fig. 4: Model development and validation.

From: Tailoring industrial enzymes for thermostability and activity evolution by the machine learning-based iCASE strategy

Fig. 4

a Workflow for dynamic response predictive model building, verifying, and evaluating. b Performance of different methods for ten datasets. c The VIP values of different variables (βT, DSI, RMSD, Rg, SASA, hydrophobicity, and charge) for ten datasets. d Performance of the dynamic response predictive model based on the random forest model for ten datasets with different data sizes. e Performance of high N datasets (beta-lactamase and PABP (RRM domain) singles) for training data sizes 48, 96, 144, 192, and 240. f The interpolation results for ten datasets. g The extrapolation results of two datasets (PABP (RRM domain) and protein-glutaminase) by training on single substituted variants and predicting higher substituted variants. The training datasets of PABP (RRM domain) and protein-glutaminase were 1188 and 45, and the test datasets were 2000 and 35. Source data are provided as a Source Data file.

Back to article page