fig1

Figure 1. (A) A graphic illustration of the FPS in chemical space; (B) A Flow chart of FPS Procedure; (C) Sampling process used in this study. The initial dataset was partitioned into training and test sets randomly. The training set was then further bifurcated into a “sampling set” and a “rest set” of different ratios by sampling using different strategies. Various ML models were next trained on the sampling set and subsequently validated on the test set. FPS: Farthest point sampling; ML: machine learning.