Fig. 3: Results of the iterations of the machine learning drug discovery approach.
From: Discovery of potent inhibitors of α-synuclein aggregation using structure-based iterative learning
a, Normalized t1/2 for the potent leads at 25 µM from the different stages: loose search, iteration 1, iteration 2 and iteration 3 (n = 2 replicates; central measure, mean; error, standard deviation). The horizontal dotted line indicates the boundary for potent lead classification, which was normalized t1/2 = 2. For the loose search, 69 molecules were tested, while for iterations 1, 2 and 3, the number of molecules tested was 64, 64 and 56, respectively. Note that the most potent molecules exhibited complete inhibition of aggregation over the timescale observed, so the normalized t1/2 is presented as the whole duration of the experiment. b, Flow of potent molecules (+) and negatives (−) in the project starting from the close search (CS), moving to the loose search (LS) and then iterations 1, 2, and 3 (I1, I2 and I3). Each branch is labeled with the molecule source (for example, p48). Attrition reached its highest point at the loose search before gradually improving with each subsequent iteration.
