There are some on-line services(e.g. molinspiration.com) that capable of prediction of logP and polar surface area for small molecules. What kind of biases can I expect from such predictions? Do they suitable for comparisons of chemically distinct molecules (e.g. cellular metabolites)? Is there any alternative for molinspiration (offline, batch) for such services converting SMILES to logP? Does anybody know the comparative study of such algorithms?
Some predicrions are very effective. For low molecular compounds it is much easier comparing to high mol. compounds. If you know some properties of some componds from the same homolog. group it is even very simple. Some properties give even linear correlation.
REGARDS
G
Hi Dmitry,
Check the following links:
https://www.organic-chemistry.org/prog/peo/
http://www.swissadme.ch/
http://fafdrugs3.mti.univ-paris-diderot.fr/
https://ochem.eu/home/show.do
http://www.vcclab.org/lab/alogps/
In general: the simpler the descriptor, the more accurate it's likely to be for a random molecule. Calculations such as logP are trained with datasets - the more similar your molecule is to the training data, the more likely it is to be inside the model's parameters, and the more accurate the output should be. For simpler descriptors (logP again is a great example, and polar surface area should be similar), they've been very well studied, and just about any reasonable small molecule should be predicted fairly accurately - but if you look at multiple logP calculators, you'll get different numbers from each, so clearly at LEAST N-1 of them are giving you data different than an experimental output. Still, they'll probably give you a good feeling for 'water soluble' vs. 'not water soluble'. Similarly, the rank-ordering of molecules within any given algorithm is probably reasonable - but it's useless to compare Bob's logP on molecule 1 to Cindy's logP on molecule 2 to Doug's logP on molecule 3.
More complex descriptors, such as ability to inhibit the enzymatic activity of some protein, or rate of metabolic degradation? If you're not fairly similar to molecules already in the dataset, the prediction is probably close to worthless.
You might also look into LICSS - it's an excel plugin that handles a bunch of different descriptor calculations, including as I recall at least one logP calculation. Chemdraw/Chemfinder will do so as well, and also plugs in to excel, but costs rather a lot more if you don't already have it available. QikProp is another good option if you happen to have access to it, but again is a pricey piece of software.
- Badges
- Science topic
- Similar topics
- Biological Science
- Botany
- Pharmaceutical Botany
- Plant Extracts