Fig. 3: The impact of multiscale conformational learning module.

Each statistic in the figure is the average of the results of ten runs using different random seeds. a Using one atom as a benchmark, we select 30% and 60% of its 2D distance to calculate the receptive field. b Results for eight of the datasets (BACE, BBBP, ClinTox, Tox21, CHEMBL1862_Ki, CHEMBL204_Ki, CHEMBL231_Ki and CHEMBL233_Ki)^36,42 using a single threshold. For each dataset, the experiment was repeated with \(n\)=10 different seeds, and the resulting figures were plotted. Blue and red colors represent the selection of receptive field based on 2D and 3D distance, respectively. The colored area represents the data distribution. For the four curves on the left, the x-axis represents the AUC value, with higher values indicating better performance. For the four curves on the right, the x-axis represents the RMSE value, with lower values indicating better performance. c Results using two thresholds on eight datasets (BACE, BBBP, ClinTox, Tox21, CHEMBL231_Ki, CHEMBL233_Ki, CHEMBL244_Ki and CHEMBL287_Ki)^36,42. Each subplot shows the effect of combining distance bars across different datasets. Darker colors represent higher values. The horizontal axis represents thresholds ranging from 20% to 100%, while the vertical axis represents thresholds from 10% to 90%. The receptive field corresponding to the point with the best value is labeled by diamonds. The mean value of the results for each row and column are given by the bar plots. d Results using three thresholds on four datasets (BACE, ClinTox, CHEMBL231_Ki and CHEMBL233_Ki)^36,42. Each plot contains \(n\)=120 sample points, ensuring the condition receptive field 1 <receptive field 2 <receptive field 3. Each point represents the mean result over ten seeds. Red-labeled points indicate optimal values, with their projected coordinates marked by dashed lines. e A comparison of using fixed thresholds versus percentage-based thresholds on four datasets (BACE, ClinTox, CHEMBL231_Ki and CHEMBL233_Ki)^36,42.1, 2, and 3 represent visual distances as spatial distances 1, 2, 3. Percent represents thresholds calculated from our percentages. Data are presented as mean ± standard deviation. Source data are provided as a Source Data file.