GenerateMD killer molecule (boran hairnet)

User 677b9c22ff

26-11-2006 01:02:27

Hi, usually GenerateMD has a throughput of 1000 molecules per

second on a Pentium M 1.7 GHz but this killer boran hairnet will

get stuck for 21 minutes in GenerateMD. Its known since 99 in the

NCI99 DB. This is the beauty:

This will also affect, Jcman, cxcalc and Instant-JChem.

The SMILES code is:


The command is:

generatemd c killer-boran.smi -k CF >garstiges-haarnetz-boran-fp.txt

The output is:


<?xml version="1.0" encoding="UTF-8"?>\n<ChemicalFingerprintConfiguration Version="0.3" schemaLocation="cfp.xsd">\n<Parameters Length="1024" BondCount="7" BitCount="3"/>\n<ScreeningConfiguration>\n <ParametrizedMetrics>\n <ParametrizedMetric Name="Tanimoto" ActiveFamily="Generic" Metric="Tanimoto" Threshold="0.2"/>\n <ParametrizedMetric Name="Euclidean" ActiveFamily="Generic" Metric="Euclidean" Threshold="10"/>\n </ParametrizedMetrics>\n</ScreeningConfiguration>\n<StandardizerConfiguration Version="0.1"><Actions><Action ID="dehydrogenize" Act="dehydrogenize"/><Action ID="aromatize" Act="aromatize"/></Actions></StandardizerConfiguration></ChemicalFingerprintConfiguration>

1 -2141187584 8456328 1073807552 134365699 67649536 16777728 50397184 135270416 2097159 2048 -1610603512 4196356 -2130702336 3163136 -2143269887 2629632 16785472 1346375680 2425920 -1073479103 0 69206050 16384 2097156 -1609023488 1074020352 109051924 285212676 1091567616 23330820 4286464 1090556928

(I know, its not a bug, its a feature :-)


ChemAxon efa1591b5a

27-11-2006 08:53:47

Hi Tobias,

thank you for the "feature-report"! ;-)

See discussion at

I have an idea how to make Fp generation faster for such monster structures even for extreme fp parameter values (such as path length = 19), I will discuss it with the developer of fp generaration and I'll get back to you afterwards.

Best regards, Miklos

Ps: Keep experimenting, it's very useful for us to see how our systems behave at such extreme conditions.

User 677b9c22ff

27-11-2006 09:54:20

Hi Miklos,

this looks like the travelling salesman when he was visiting Königsberg.

Looking at this nasty/beauty molecule (depending on the point of view - developer/user) -

maybe a genetic algorithm with simulated annealing can help. Or maybe using Bayesian Networks

and the Max-Min Hill-Climbing Algorithm [PDF]. But now I am happy, I am just a user .-)