I have a large set of molecules and I'd like to assess its diversity in terms of molecular structure, assuming I'm designing a screening library. My question is not what tools to use, but rather how should I interpret the results?
Let's say I choose clustering. How many clusters indicate a diverse library?
"Modern Approaches in Drug Discovery" suggests comparing Morgan fingerprints against Tanimoto distance matrix with multidimensional scaling applied, among others. But what will it get me?
Dear Anna,
Please find sources/examples below, maybe some will be useful for you,
Article Design of a multi-purpose fragment screening library using m...
Article Chemistry Space Metrics in Diversity Analysis, Library Desig...
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5105260/
Article Scaffold Diversity of Fungal Metabolites
http://infochim.u-strasbg.fr/CS3/program/material/Gillet.pdf
Article Cheminformatics approaches to analyze diversity in compound ...
Pozdrowienia
Tomasz
Here is one way to assess the diversity of a library:
1. Find out an approximate similarity threshold for the metric/descriptors used in the diversity analysis. For example, for the popular MACCS keys, the Tanimoto threshold is ~0.75; for ECFP4 (circular fingerprints implemented in Pipeline Pilot), the Tanimoto threshold is ~0.5.
2. Calculate all pairwise similarities for your library compounds.
3. Plot and analyze the probability density function for these similarity values. An ideally diverse library would not feature any pairwise similarity higher than the threshold (that is, every molecule is dissimilar from each other). For a fully redundant library, all pairwise distances would be higher than the threshold (that is, every molecule is similar to each other).
4. It is you call then to decide which kind of library do you need? That is, how much of redundancy would you accept? Keep in mind that an ideally diverse library is not well suited for screening. Ideally, every compound would be a member of a small family (cluster) of 10-20 compounds. If more than one family members show some activity in a screen, this may be considered as an indirect confirmation that these are true actives. Also, activity values for the family members can be used for a preliminary structure-activity relationships (SAR) analysis.
Dmitri Kireev , what is the point of calculating probability density function in this case?
- Badges
- Science topic
- Similar topics
- Chemistry