The drift is pulling ligands away from established binding sites.

[electronicsbyjulie]

This may be a tricky issue to solve.

When using a path based dock, there is a drift that nudges the ligand towards the current path node. If the ligand has formed a bond with the receptor, that bond can be pulled apart. This is evident when docking either of the diamines in TAAR6 or TAAR8; at each step, there are two side chains available for ionic bonding or hydrogen bonding, and one of the side chains was also part of the previous node. Each amino group should stay near the shared side chain, perhaps dragging the moiety along causing the side chain to flex, while the other amino group joins up to the newly selected side chain. Effectively, the ligand should "walk" into the binding pocket by having its two strongest moieties "leapfrog" over each other.

To solve this in a way that works for all possible ligands is key. The algorithm should seek to identify the shared strongly bound residue and perform a rotation of the ligand about the residue's alpha carbon, along with a conformation of the side chain to keep it optimized to the ligand. But in cases where there is no strong moiety, such as a thiol traveling through OR2T11, or a fatty alcohol or a muguet aldehyde moving deeper into a binding pocket, then no such "walking" rotation will be possible, and the current drift behavior would be preferred.

Knowing the axis of rotation might depend on measuring the distances between moieties of the molecule and correlating a molecular feature with the newly selected side chain. The possibility exists of interference during the motion from unrelated side chains that happen to block the path. Those may influence the likelihood of the "walk" step taking place at all, and/or may cause it to go around the obstacle at a modified angle. All of these things would affect the energy probability of the real world receptor-ligand system.

Ultimately, the real world molecules are following vast numbers of paths every second, and the docker can only hope to estimate the path with the greatest chance of receptor activation. The statistical probability of this path will also affect the final result - i.e. the ligand is or is not an agonist, and if so, how strong of one - but the computation required to estimate this probability may require extremely long processing times under the capabilities of machines of the present time.

[primaryodors]

You have the atom of the receptor that is most strongly bound to the ligand. We'll call its xyz coordinates A for atom. Then you have the barycenter of the ligand, we'll call it B for barycenter. Finally, there's the target node center, let's call it C for center.

Take the normal of ABC, call it N. Then you can rotate the ligand about A axis N to bring B closer to C in lieu of a drift.

[electronicsbyjulie]

Working in the ssepeq-leapfrog branch, it has given a correct prediction for TAAR8 cadaverine, and the ligand is leapfrogging along the path of acid groups, but there are still some ionic bonds coming undone. Including salt bridges in the protein structure such as Arg91-Glu293.

[electronicsbyjulie]

152 addresses this, but established ionic bonds are still coming loose so leaving the ticket open.

[primaryodors]

Let's check whether this is still happening.