I'm fairly new to the field of cheminformatics so I would be very grateful for experienced suggestions. I have three question (at the bottom), followed by two further questions that I thought of later.
Firstly, I've been making some fantastic progress with a pilot study looking at 2D structural similarity (encoded across a number of different fingerprints and similarities calculated across a number of different methods) and drug target predictions, all backed up with a large in vitro assay.
However, I'm very keen to investigate whether I find similar results/patterns if I was to consider the drug's 3D structure. I've currently been using RDKit to implement a high-throughput 2D similarity screening and from what I can find, the only 3D descriptors in RDKit are Asphercicity, Eccentricity, InertialSphapeFactor, NPR1, ..., PM1, ..., RadiusOfGyration and SpherocityIndex. I can expand this using the Mordred python API (1800+ descriptors). However, if I am correct, these are all descriptors, not actual 3D structural fingerprints.
1) Is there a 3D structure fingerprint encoder in RDKit with a means of similarity calculation? e.g., Tanimoto, Dice, etc, between two 3D structures?
2) If there is something available, what are people's experiences with using it? I've read accounts that generating a 3D structure from a SMILES can run into problems especially when chirality is considered.
3) If there is nothing suitable in RDKit, are there any recommendations for (a) free software for academia, (b) can be incorporated into a pipeline e.g., a python API available?
As a follow-up to my earlier question (above).
1) I found an uncited mention of HOMO and LUMO energy levels being used as molecular descriptors. This sounds fascinating! Sadly, my Google skills clearly aren't up to the task. If anyone has any information on this technique I would be very interested to learn more.
2) On a similar note, is anyone aware of theoretical IR spectra being used as a means of calculating the similarity between compounds in a drug-screening style/high-throughput fashion?
Check out Structural Protein-Ligand Interaction fingerprints (SPLIF), available as a part of the DeepChem suite.
https://github.com/deepchem/deepchem/issues/876
Article Structural Protein-Ligand Interaction Fingerprints (SPLIF) f...
Dmitri Kireev many thanks - it's printed. Now I've just got to find the time to read it!
Isomeric SMILES are unambiguous and can be converted into 3D structures using molconvert (ChemAxon).
https://docs.chemaxon.com/display/docs/molecule-file-conversion-with-molconverter.md
- Badges
- Science topic