deletions | additions       diff --git a/subsection_Cathepsin_G_Virtual_screening__.tex b/subsection_Cathepsin_G_Virtual_screening__.tex  index fe046c9..cd01aeb 100644  --- a/subsection_Cathepsin_G_Virtual_screening__.tex  +++ b/subsection_Cathepsin_G_Virtual_screening__.tex 
...
\subsection*{Cathepsin G}  Virtual screening results for cathepsin G are shown in Figure~\ref{cathg}. All three shape similarity methods exhibited poor performance with this query, query with AUCs near 0.5,  but many of the interaction point queries were significantly better than random. The statistically most significant interaction point query is shown in Figure~\ref{cathgip}. The fragment for the query is deeply buried within the S1 pocket and the interaction points match the S2 pocket and phosphonate moiety of the ligand. However, there are no interaction points selected in the S3/S4 pocket. Interactions in this pocket can increase affinity, but are not necessary for binding \cite{de_Garavilla_2005}. Unlike whole-ligand similarity methods, shape constraint search can ignore a significant part of the geometry of the ligand (the naphthalene in the S3 pocket) and select for only specific parts of the query shape. In this case, this selectivity results in substantially better virtual screening results that is two orders of magnitude faster than they pre-aligned similarity search methods.