Balancing Data on Deep Learning-Based Proteochemometric Activity Classification.

Figure 4 puts the predicted ratios in the context of the training ratios, and Table S8 of the Supporting Information quantifies their correlations, elucidating a variety of trends: (1) no_resampling shows a positive trend between the training and the predicted ratio (Pearson’s r 95% CI: [0.440, 0.496], p < 10 –16 ), but since the training and the test ratio are also positively correlated ( Figure 3 ), the latter could be the one driving the predicted ratio of positives; (2) resampling_after_clustering had a constant

also in 11,005 other papers