Figure 6. Re-evaluation of our method using the data of experimentally verified chemical compounds.

(A) Evaluation by recall rate with 10,000 chemical compounds. Here, the recall rate at the rank x in descending order of predicted binding probability was calculated as (the number of 15 binding ligands whose rank is higher than x)/15. The 10,000 tested chemical compounds included 1,041 predicted ligand candidates, as shown in Figure 1, and 8,959 of the compounds were found within PubChem Compound CID 1-10427. (B) AUC using the data of 15 experimentally verified ligands and 8 non-binding chemical compounds. ^*: In both (A) and (B), the prediction models were constructed as described in Figure 4, where 6 known chemical compound-androgen receptor pairs or 6 known chemical compounds among the 23 verified chemical compounds were excluded from the dataset utilized to construct the final prediction model and the weighting factor for two-layer SVM models was set to 10.