Smiles String Comparison Algorithms

Question

[Biogeek]

What are the similarity algorithms normally used to compare slightly different, but related SMILES strings (e.g. Oc1ccc(cc1)\C=C\C(=O)c2ccc(O)cc2O vs O=C(/C=C/c1ccccc1)c2ccccc2).

Answer 1

[brentp]

See this by Andrew Dalke.

In it, he references:

Lingos, Finite State Machines, and Fast Similarity Searching", J. A. Grant, J. A. Haigh, B. T. Pickup, A. Nicholls, and R. A. Sayle, J. Chem. Inf. Model 46(5) (2006) p1912-1918.

He also looks at using compression via zlib to look at compression.

Answer 2

[Egon Willighagen]

Comparing SMILES directly only makes sense when you use canonical SMILES. More common is to process the SMILES in a chemical graph, and compare the actual graphs, so that it does not matter that you can have multiples SMILES for the same molecule. From then on, I suggest the fingerprint as representation for which you can calculate the similarity with the Tanimoto distance.

Example code using the CDK and R can be found in this vignette using the rcdk package.

[Noel O'Boyle]

To expand...there can be many SMILES strings for the same chemical structure, so it doesn't make sense to compare the strings themselves.

Answer 3

[Gilleain]

You can use the SMSD to compare molecules as SMILES, which gives various similarity measures including Tanimoto.