Menu ▾ ▴

#1195 some descriptors fail to handle aromaticity correctly

open
nobody
cdk-qsar module (68)
9
2012-10-08
2011-12-21
Nikolay Kochev
No

A set of descriptors does not work correctly for the two types of aromaticity representation e.g. c1ccccc1 and C1=CC=CC=C1 give quite different result values for the descriptor.
Here are listed 3 examples and the the list of all descriptors that fail:

Comparing descriptor values for c1ccccc1 and C1=CC=CC=C1
XLogP 4.062 2.082
C2SP2 0.0 6.0
C2SP3 6.0 0.0
ALogP 0.0 1.4033999999999995
ALogp2 0.0 1.9695315599999987
AMR 0.0 30.955799999999996
BCUTc-1l -0.18066486607184457 -0.21175815206924314
BCUTc-1h 0.11938953556395132 0.08829624956655291
MolIP 0.0 8.7360121

Comparing descriptor values for c1ccccc1CCC(CC)CCN and C1=CC=CC=C1CCC(CC)CCN
LipinskiFailures 1.0 0.0
XLogP 5.906 3.8899999999999997
ATSc1 0.02963300936257801 0.030880312144664857
ATSc2 -0.012090061482428298 -0.012341625539837791
ATSc3 -0.0025292327510609384 -0.0026466176367462594
ATSc4 -1.8482519638761126E-4 7.312021157925057E-4
ATSc5 -1.2229491464390043E-5 0.001777028789656713
C2SP2 0.0 5.0
C3SP2 0.0 1.0
C2SP3 9.0 4.0
C3SP3 2.0 1.0
ALogP -1.2510999999999997 -0.04330000000000056
ALogp2 1.5652512099999991 0.0018748900000000485
AMR 32.7762 62.511900000000004
BCUTc-1h 0.027581097463993814 0.027504421178008526
MolIP 9.78147624160733 8.199589395397101

Comparing descriptor values for c1ccccc1CCC(CC)CCNc2ccccc2 and C1=CC=CC=C1CCC(CC)CCNC2=CC=CC=C2
XLogP 8.838999999999999 5.454000000000001
ATSc1 0.062282611216200066 0.05057222131722536
ATSc2 -0.03115211769438775 -0.025073172136074226
ATSc3 -0.004660620187309899 -0.003476970136158749
ATSc4 0.003572779513132754 0.003487608390774221
ATSc5 9.70815694659988E-4 0.0018642068852909237
C2SP2 0.0 11.0
C3SP2 0.0 1.0
C2SP3 15.0 4.0
C3SP3 2.0 1.0
ALogP -0.9215999999999995 1.2428000000000012
ALogp2 0.8493465599999991 1.5445518400000031
AMR 31.8541 91.87329999999999
BCUTc-1l -0.30846158469250606 -0.2884631663210181
BCUTc-1h 0.16905183889116193 0.13855538667368522
MolIP 8.496508028720328 5.114511566145805

And here is given the structure pre-processing

void processStructure(IAtomContainer ac) throws Exception
{
clearAromaticityFlags(ac);
AtomContainerManipulator.percieveAtomTypesAndConfigureAtoms(ac);
CDKHydrogenAdder adder = CDKHydrogenAdder.getInstance(SilentChemObjectBuilder.getInstance());
adder.addImplicitHydrogens(ac);
//AtomContainerManipulator.convertImplicitToExplicitHydrogens(ac);

    CDKHueckelAromaticityDetector.detectAromaticity(ac);
}

Discussion

  • Nikolay Kochev

    Nikolay Kochev - 2011-12-21

    In my opinion this bug is serious since the descriptors are used for very important tasks. For example XLogP gives the 'better' value (2.082) for C1=CC=CC=1 representation and quite erroneous value (4.062) for c1ccccc1 representation of benzene. The experimental LogP value for benzene is 2.13.
    This is a big problem because majority of people would prefer the aromatic form instead of Kekule one. It seems that this problem can be seen in many descriptors.

     

  • Egon Willighagen

    Egon Willighagen - 2012-08-02

    Agreed. This is important. If that preprocessing has taken place, these descriptors should give the same values, no matter what SMILES input was given...

    Since the various descriptors have different authors, I suggest to split up this bug report in separate reports for the various descriptor classes...