Fig. 2

Comparison of model performance for CYP2B6 (A F1, B Cohens-Kappa) and CYPC8 (C F1, D Cohens-Kappa). Single: single-task learning/baseline; Multi: multitask learning based on CYP isoform data; FT_CYP isoform: fine-tuning models based on CYP isoforms; MIPS: multitask learning with the imputation of predicted label from single-task model; MIPM: multitask learning with the imputation of predicted label from multitask model; MIPFT: multitask learning with the imputation of predicted label from fine-tuning model. *Significant compared to single-task model (p < 0.05), **Significant compared to single-task model (p < 0.01)