Figure 3 | Scientific Reports

Figure 3

From: Studying and mitigating the effects of data drifts on ML model performance at the example of chemical toxicity data

Figure 3

Analysis of individual endpoints (a) Balanced evaluation of time-split experiments for four selected ChEMBL endpoints. Each plot represents CV results (cv) and predictions for the holdout set using the original (cal_original), update1 (cal_update1), update2 (cal_update2) and combined update1_and_2 (cal_update1_and_2) calibration sets. The doted line at 0.8 denotes the expected validity for the chosen significance level of 0.2. (b) UMAP showing the descriptor space covered by the compounds in the different time-split sets for ChEMBL206 endpoint.

Back to article page