Article ID: 172 - Last Modified: December 4, 2010
How good is QikProp in real world usage?
QikProp calculates a number of properties with varying degrees of real world usefulness.
Obviously properties which can be directly calculated from the molecular formula or the molecular structure will be highly accurate or exact. These include properties such as MW, rotatable bonds and surface areas.
More complex properties will depend critically on the data used to train the model and the nature of the property being calculated. For example, the literature contains a large range of experimentally determined logP(o/w) values and it is possible to generate a highly reliable model of what is essentially a simple physical process.
Solubility however is far more complex. QikProp, like many solubility models, is trained on a dataset which spans a very wide range of solubilities (around 7 orders of magnitude) and which covers a huge range of different chemical structures. However, in real world usage it is rare to need to consider such a large variation in solubility and structure. Instead it is more normal to consider small variations around a given core. Unfortunately predicting these subtle variations accurately is extremely hard, as they are dependent on a huge number of essentially unknown parameters about the nature of the solid state and the ease of liberating molecules from the lattice.
The most complex predictions, properties such as HERG blockage, BBB penetration and CaCO2 permeation should be considered as vague estimations only. Accurately modeling any of these parameters, particularly when the initial data set is so small and variable, is extraordinarily hard.
Type the words or phrases on which you would like to search, or click here to view a list of all
Knowledge Base articles