SMARTS definitions for interactions

[cbouy]

As mentioned in issue #68 and PR #73, the SMARTS patterns used in ProLIF are too simple in some cases and can incorrectly detect or miss some interactions.

There are already a couple of suggestions from @DrrDom on how to improve those, but the discussion is still open:

HBondAcceptor

Change acceptor SMARTS ([N,O,F,-{1-};!+{1-}]) to a more comprehensive one - [#7&!$([nX3])&!$([NX3]-*=[!#6])&!$([NX3]-[a])&!$([NX4]),$([O])&!$([OX2](C)C=O)&!$(*(~a)~a),-{1-};!+{1-}] (the source of some patterns is here https://sourceforge.net/p/pharmit/code/ci/master/tree/src/). This will exclude some N and O atoms.

MetalAcceptor

Ligand [O,N,-{1-};!+{1-}] will match only aliphatic N and O, at least aromatic N can interact with a metal center. So, it might be better to fix it - [O,#7&!$([nX3])&!$([NX3]-*=[!#6])&!$([NX3]-[a])&!$([NX4]),-{1-};!+{1-}].

Hydrophobic

Hydrophobic interaction pattern [#6,#16,F,Cl,Br,I,At;+0] is too broad, it includes tetracoordinated S and C atoms which are not accessible to any interaction and will result in false positive detection. It can be fixed like [$([#6])&!$([#6X4H0]),$([#16])&!$([#16X3])&!$([#16X4]),F,Cl,Br,I;+0]. Here I excluded S with 3 connections as well (sulfoxides). It might be also reasonable to exclude C in carboxylates.

A source of inspiration could also be the feature definitions used by RDKit for 2D pharmacophores and Morgan fingerprints.