So much to do, so little time

Trying to squeeze sense out of chemical data

Archive for the ‘fingerprint’ tag

Updates to R Packages

without comments

I’ve uploaded a new version of fingerprint (v 3.4) which now supports feature fingerprints – fingerprints that are represented as variable length vectors of numbers or strings. An example would be circular fingerprints. Now, when reading fingerprints you have to indicate whether you’re loading binary fingerprints or not (via the binary argument in A new line parser function (ecfp.lf) is provided to load these types of files, though it’s trivial to write your own. Similarity can be evaluated between feature fingerprints in the usual manner, but the metrics are restricted to Tanimoto and Dice. A function is also available to convert a collection of feature fingerprints into a set of fixed length binary fingerprints ( as described here.

New versions of rcdk (v 3.0.4) and rcdklibs (v have also been uploaded to CRAN. These releases are based on todays CDK 1.4.x branch and resolve a number of bugs and add some new features

  • Correct formula generation
  • Correct handling of SD tags whose values are just white space
  • Proper generation of Murcko frameworks when molecule objects are requested
  • 3 new descriptors – FMF, acidic group count, basic group count

Written by Rajarshi Guha

October 22nd, 2010 at 1:58 am

Posted in cheminformatics

Tagged with , ,

A Comment on Fingerprint Performance

without comments

In a comment to my previous post on bit collisions in hashed fingerprints, Asad reported on some interesting points which would be useful to have up here:

Very interesting topic. I have faced these challenges while working with fingerprints and here are few observations from my end. By the way I agree that mathematically the best bet is ~ 13%.

1) The hashed FP (CDK) is good enough to separate patterns which are not common but on a large dataset (in my case 10000+ mols), the performance drops drastically. Top 1% hits were good but then rest of the started to loose specificity (esp when Tanimoto score was around 0.77).

2) First I thought it was an artifact of the Tanimoto score… but I wasn’t convinced spl. in cases where we had rings (close vs open). I ended up writing a new FP based on the pubchem patterns as coded in the CDK and added few more patterns to resize it to 1024 from 881. Well! It’s works like magic and I could find much more serialised hits than before. I think the extensions of the fingerprint which I made based on the patterns in my db also helped.

At the end of the day, I believe that all these searches are heuristic and hashed FP is faster to generate but prone to bit clashes where as SMARTS based FPs are slower to generate (as u spend time in MCS) in matching patterns but they are more sensitive and specific as you can trace the patterns (u get what u see) as the patterns and bitset relationship is know and static

Written by Rajarshi Guha

October 9th, 2010 at 3:23 am

Posted in cheminformatics,software

Tagged with , ,

Hashed Fingerprints and RNG’s

with 10 comments

In my previous post I looked at how many collisions in bit positions were observed when generating hashed fingerprints (using the CDK 1024-bit hashed fingerprint and the Java hashCode method). I summarized the results in the form of “bit collision plots” where I plotted the number of times a bit was set to 1 versus the bit position (for a given molecule). As expected, for a series of molecules we observe a number of collisions in multiple bits. What was a little surprising was that even for a symmetric molecule like triphenylphosphine (i.e., a relatively small number of topologically unique paths), we observed collisions in two bits. So I decided to look into this case in a little more detail.

As I noted, collisions could occur if a) different paths get hashed to the same int or b) two different hashes lead to the same random number. Modifying the Fingerprinter code, I was able to generate the list of paths calculated for triphenylphosphine, the hash code for each path and the bit position that was generated for that hash code. The data (hash value, path, bit position) is given below.

-662168118  P-C:C-H 3
-1466409134 H-C:C-P-C:C:C   58
1279033458  C:C:C:C:C-P-C:C:C   78
1434821739  H-C:C:C:C:C:C   80
-429128489  C:C:C:C-P-C 85
-1779215129 C:C:C:C:C   95
-245205916  H-C:C:C-P-C:C:C:C   97
-475263438  C:C:C:C:C:C-P-C:C   111
-1466409532 H-C:C-P-C:C-H   114
1434821341  H-C:C:C:C:C-H   142
-428753296  C:C:C:C:C:C 161
-245206314  H-C:C:C-P-C:C:C-H   161
1730724873  H-C:C:C-P-C 167
43327460    H-C:C:C:C:C-P-C:C   180
1731099668  H-C:C:C:C-H 180
886716569   H-C:C:C:C   182
-1406488727 C:C:C:C:C-P-C:C 191
78342   P-C 193
1731100066  H-C:C:C:C:C 211
63670037    C:C:C   213
178815835   H-C:C:C:C:C-P-C 224
179190630   H-C:C:C:C:C:C-H 230
-469902821  H-C:C-P-C:C:C:C 244
1057365278  C:C:C:C 253
-181369445  H-C:C:C:C-P-C:C 266
1056990085  C:C-P-C 300
886341376   H-C:C-P-C   333
827737424   H-C:C:C 390
-912402715  P-C:C:C:C:C-H   406
1572927053  H-C:C:C-P-C:C-H 421
1434446546  H-C:C:C:C-P-C   440
403512740   H-C:C:C:C:C:C-P-C   442
827737026   H-C:C-H 448
1572927451  H-C:C:C-P-C:C:C 458
-645296786  P-C:C:C:C:C:C-H 493
-688017645  P-C:C:C-H   503
-763796814  C:C:C:C-P-C:C:C:C   512
66252   C:C 574
75288527    P-C:C   600
-605043036  H-C:C-P-C:C:C:C:C   604
1074261266  H-C:C:C-P-C:C   629
-789313769  C:C:C-P-C:C 639
-2139775602 C:C-P-C:C   675
1370539593  H-C:C-P-C:C 698
284569632   C:C:C:C:C-P-C   702
1797624356  H-C:C:C:C-P-C:C:C   710
63294844    C-P-C   719
-912402317  P-C:C:C:C:C:C   725
72  H   741
67  C   742
80  P   744
-1779590322 C:C:C-P-C   762
1797623958  H-C:C:C:C-P-C:C-H   774
347808169   C:C:C:C-P-C:C:C 788
-688017247  P-C:C:C:C   815
240390684   P-C:C:C:C-H 834
-75615648   C:C:C:C-P-C:C   859
240391082   P-C:C:C:C:C 866
886716171   H-C:C:C-H   888
67900359    H-C:C   951
-1046303447 C:C:C:C:C:C-P-C 957
-662167720  P-C:C:C 969
70654   H-C 971
1678681248  C:C:C-P-C:C:C   979

First, all the hash codes are unique. So clearly the issue lies in the RNG and indeed, we see the following two paths being mapped to the same random integer.

-428753296    C:C:C:C:C:C       161
-245206314    H-C:C:C-P-C:C:C-H 161

Does this mean that the two hash values, when used as seeds to the RNG give the same sequence of random ints? Using the code below

Random rng1 = new Random(-428753296);
Random rng2 = new Random(-245206314);
for (int i = 0; i < 5; i++) {
    System.out.println(rng1.nextInt(1024) + " " + rng2.nextInt(1024));

we generate the first five random integers and we see that they match at the first value but then differ.

161 161
846 40
317 885
461 535
448 982

This suggests that instead of using the first random integer from the RNG seeded by a hash value, we use the second random integer. Modifying the code to do this still gives collisions in two bits. Once again, looking at the paths, hashes and bit positions, we see that now, two different paths get mapped to the same bit position.

886341376      H-C:C-P-C      686
1434821341     H-C:C:C:C:C-H  686

As before, we look at the sequence of random ints obtained from RNG’s seeded using these hash values. The resultant sequence looks like:

333 142
686 686
905 1022
70 571
177 384

So now, the two sequences match at the second value. OK, so what happens if we take the third value from the sequence and use that as a bit position? We get exactly the same behavior (collisions at two bit positions), except that now, when we look at the sequence of random int’s they match at the third value.

This behavior seems a little strange to me – as if there is a pair of seeds such that the “trajectory” of the sequences generated using those seeds will always (?) intersect at a certain point (where point actually corresponds to the n’th element of the sequences).

May be this is a property of random sequences? Or a feature of the Java RNG. I’d love to hear if anybody has insight into this behavior.

Written by Rajarshi Guha

October 4th, 2010 at 7:30 am

Path Fingerprints and Hash Quality

with 5 comments

Recently, on an email thread I was involved in, Egon mentioned that the CDK hashed fingerprints were probably being penalized by the poor hashing provided by Java’s hashCode method. Essentially, he suspected that the collision rate was high and so that the many bits were being set multiple times by different paths and that a fraction of bits were not being touched.

Recall that the CDK hashed fingerprint determines all topologically unique paths upto a certain length and stores them as strings (composed of atom & bond symbols). Each path is then converted to an int via the hashCode method and this int value is used to seed the Java random number generator. Using this generator a random integer value is obtained which is used as the position in the bit string which will be set to 1 for that specific path..

A quick modification to the CDK Fingerprinter code allowed me to dump out the number of times each position in the bitstring was being set, during the calculation of the fingerprint for a single molecule. Plotting the number of hits at each position allows us to visualize the effectiveness of the hashing mechanism. Given that the path strings being hashed are unique, a collision implies that two different paths are being hashed to the same bit position.

The figure alongside summarizes this for the CDK 1024-bit hashed fingerprints on 9 arbitrary molecules. The x-axis represents the bit position and the y-axis on each plot represents the number of times a given position is set to 1 during the calculation. All plots are on the same scale, so we can compare the different molecules (though size effects are not taken into account).

Visually, it appears that the bit positions being set are uniform randomly distributed throughout the length of the fingerprint. However, the number of collisions observed is non-trvial. While for most cases, there doesn’t seem to be a significant number of collisions, the substituted benzoic acid does have a number of bits that are set 4 times and many bits with 2 or more collisions.

The sparsity of triphenyl phosphine can be ascribed to the symmetry of the molecule and the consequent smaller number of unique paths being hashed. However it’s interesting to note that even in such a case, two bit positions see a collision and suggests that the hash function being employed is not that great.

This is a quick hack to get some evidence of hash function quality and its effect on hashed fingerprints. The immediate next step is to look at alternative hash functions. There are also other aspects of the measurement & visualization process that could be tweaked – taking into account molecular size, the actual number of unique paths and converting the plots shown here to some concise numeric representation, allowing us to summarize larger datasets in a single view.

Update – I just realized that the hash function is not the only factor here. The Java random number generator plays an important role. A quick test with the MD5 hash function indicates that we still see collisions (actually, more so than with hashCode), suggesting that the problem may be with how the RNG is being seeded (and the fact that only 48 bits of the seed are used).

Written by Rajarshi Guha

October 2nd, 2010 at 5:09 pm

SALI in Bulk

without comments

Sometime back John Van Drie and I had developed the Structure Activity Landscape Index (SALI), which is a way to quantify activity cliffs – pairs of compounds which are structurally very similar but have significantly different activities. In preparation for a talk on SALI at the Boston ACS, I was looking for SAR datasets that contained cliffs. It turns out that ChEMBL is a a great resource for SAR data. And with the EBI providing database dumps it’s very easy to query across the entire collection to find datasets of interest.

For the purposes of this talk, I wanted to see what the datasets looked like in terms of the presence (or absence of cliffs). Given that the idea of an activity cliff is only sensible for ligand receptor type interactions, I only considered compound sets associated with binding assays. Furthermore, I only considered those assays which involved human targets, had a confidence score greater than 8 and contained between 75 and 500 molecules. (If you have an Oracle installation of ChEMBL then this SQL snippet will get you the list of assays satisfying these constraints).

This gives us 31 assays, which we can now analyze. For the purposes of this note, I evaluated the CDK hashed fingerprints and used the standardized activities to generate the pairwise SALI values for each of the datasets (performing the appropriate log transformation  of the activities when required). The matrices that represent the pairwise SALI values are plotted in the heatmap montage below (the ChEMBL assay ID is noted in each image) where black represents the minimum SALI value and white represents the maximum SALI value for that dataset. (See the original paper for more details on this representation.) Clearly, the “roughness” of the activity landscape differs from dataset to dataset.

At this point I haven’t looked in depth into each dataset to characterize the landscapes in more detail, but this is a quick summary of multiple datasets. (Though a few datasets contain cliffs which are derived from stereoiomers and hence may not actually be real cliffs – since their activity difference may be small, but will look structurally identical to the fingerprint).

An alternative and useful representation is to convert the SALI values for a dataset into an empirical cumulative distribution function to provide a more quantitative view of how cliffs are distributed within a landscape. I’ll leave those details for the talk.

Written by Rajarshi Guha

August 11th, 2010 at 4:31 am