deletions | additions       diff --git a/subsection_HSP90_The_best_SMARTS__.tex b/subsection_HSP90_The_best_SMARTS__.tex  index 391b16b..acf5b2c 100644  --- a/subsection_HSP90_The_best_SMARTS__.tex  +++ b/subsection_HSP90_The_best_SMARTS__.tex 
...
\subsection*{HSP90}  Virtual screening results heat shock protein 90 (HSP90) are shown in Figure~\ref{hsp90}. All three shape similarity methods perform poorly with AUCs from 0.38 (VAMS) to 0.51 (RDKit). The shape constraint queries can produce statistically significant enrichments and the pareto frontier matches or exceeds the ROC curve of RDKit. This suggests that this target benefits from a partial similarity search as opposed to a whole molecule similarity.  %  The best SMARTS expression and approach for Hsp90 performed significantly better than random performance for FOMS, but not for VAMS (cite table). Hsp 90 inhibitors interact with the ATP binding site. The most important interactions are with Asp93, Gly97, Lys58 , and Asn51 residues, with the study reporting the crystal structure highlighting Asp93 as the most important (cite HSP). Only the Asp93 interaction is covered by our SMARTS expressions and fragment pre-alignments, while the other interactions are also key for Hsp90 inhibition, with the Gly backbone being second in importance. As with other targets in the MUV dataset, the poor performance of FOMS relative to random performance can be partially attributed to the large number of interactions that are not covered by our fragment selection due to the nature of the ligand binding mode. %  Docking with 2VCI in smina for the active compounds from the MUV dataset were performed. This revealed 4 potential hydrophobic pockets, with only one covered by our fragment pre-alignment. There were also many active conformers hydrogen-bonding with Asp93 and at least one other residue as well. These docking results indicate the importance of inclusion of every hydrogen-bonding interaction and more than one hydrophobic pocket for Hsp90 inhibition. Our SMARTS expression and fragment pre-alignment only covered one of these interactions and one of the pockets, which most likely led to FOMS not having high performance. More %More  specific SMARTS expressions led to higher performance for FOMS and VAMS (cite table). This makes sense as the more specific fragment choice would help both methods select compounds that can make the hydrogen-bond with Asp93.