Figure 4
From: Efficient discovery of responses of proteins to compounds using active learning
Active learning to discover compound-target hits. Experiments were selected which had the largest absolute predictions. The average number of discoveries and standard error for 10 separate trials are shown. The selection methods were random selection (red), CCT with greedy selection (green), greedy selection with single regression using only compound features for prediction (CFO, cyan) and greedy selection with single regression using only target features for prediction (PFO, blue).