Closed primaryodors closed 8 months ago
One cause may be that the prediction method is getting confused which PDB is the active vs. the inactive model. This discrepancy causes sporadic incorrect negative predictions for agonists.
The dock algorithm is classifying histidines as basic residues, when histidine side chains in OR51E* remain in their uncharged hydrophobic state. The use of a pH constant and pKa calculation is throwing off the amino test (see #374) and throwing off the dock. Histidine side chains should always stay uncharged and nonpolar unless hydrogen bonded to a negative charge.
Even if the histidines have incorrect charges, causing incorrect results, the results should still be self consistent. The stochasticity of the dock should converge on the theoretical most energetically favorable pose; it should not randomize the outputs.
OR51E2 is now giving consistent predictions: butyric acid has the highest dock score, followed by propionic and acetic acids, with formic and valeric acids not far behind. Isovaleric acid, a weak agonist, tends to (incorrectly) produce a dock score of zero, while diacetyl, a non-agonist, almost always has a positive score. The other non-agonists consistently produce either a score of zero or a failed dock, as do the lactone agonist and the two ionones.
OR51E1 is giving consistently incorrect predictions every time.
Therefore, while the predictions are still not fully accurate, they are now at least consistent.