High Quality like Fingerprints but Complementary Results

The following chart compares SimFinder and Fingerprints for 5 activity classes in the MDDR database. The y-axis shows the ROC-100 score (higher means more relevant matches in the top-100 results returned).

While SimFinder is obviously very competitive by itself, its true strength stems from its ability to provide complementary results to the user, resulting in a much higher chance for new leads and differentiation for drug makers.

This point can be illustrated by drilling down into the HMG result. As the following diagram shows, while SimFinder delivers at least as many actives in its search result as Fingerprints, the two result sets have a small overlap. The molecules in the left part are only found by SimFinder within the top 100 results but may proof to be valuable leads.

Meaningful Result Ranking

SimFinder operates entirely in atom/bond space without going through any bitmap conversions such as Fingerprints. This allows SimFinder to generate very high quality result ranking that is more correlated with actual activity strength than for other techniques.

For example, consider the Dopamine D2 receptor binding class from the PDSP dataset. The ranking of molecules generated by SimFinder when given the lowest energy (Ki) compound is very correlated to the ranking by energy of these molecules. This can be seen in the following plots.

The first plot shows the correlation between Ki values and ranking scores of SimFinder and Fingerprints. Clearly, SimFinder exhibits a good correlation while Fingerprints show no correlation at all. The second plot is the ROC curve for the same dataset. The x-axis shows the number of false positives (molecules with Ki value > 10) encountered and the y-axis the number of true positives (molecules with Ki value < 10). Over the entire plot, SimFinder returns more true positives, and therefore more meaningful matches, than Fingerprint techniques.