Figure 4.

Pairwise chemical similarity of predicted, random, latest PDB, and query compounds. Top: violine plot. Bottom: empirical cumulative density function (ECDF) of similarities. Query compounds are more similar to each other than predictions, which are as similar to each other as a random set of compounds. This indicates that predictions substantially expand from the queries and cover a vast chemical space. 99% of random compounds have a similarity of less than 0.25 suggesting that 0.25 is a meaningful cut-off to consider compounds unrelated.