Predicting Properties from Molecular Structure

Quantitative structure–property relationship (QSPR) models use molecular structure to predict experimentally relevant properties. In a study, researchers introduced a new edge-connectivity descriptor designed to account for heteroatoms and multiple bonds in molecular graphs.

Using this descriptor together with parameters describing the position and chemical environment of the hydroxyl group, the researchers developed regression models for five properties of alcohols:

• Normal boiling point

• Molar volume

• Molar refraction

• Aqueous solubility

• Octanol–water partition coefficient, or logP

All five models produced correlation coefficients greater than 0.99, and leave-one-out cross-validation indicated that their performance was statistically stable. The study demonstrates how chemically meaningful molecular descriptors can translate structural information into quantitative property predictions. However, the models were developed for a defined set of relatively small alcohols, so their performance should not be assumed to extend automatically to unrelated chemical classes.

Study: https://06ef7d31-235b-4052-b118-315848038251.usrfiles.com/ugd/06ef7d_ad5a313c68694eac96a68491abef1aa2.pdf

How CNDDL Can Help

Canadian Drug Discovery Lab develops customized computational models to support molecular design, virtual screening, and lead optimization. CNDDL’s in silico workflows can integrate molecular descriptors, machine learning, structural modeling, and experimental data to predict properties such as solubility, lipophilicity, and other indicators of drug developability. These models can help prioritize compounds before synthesis, reduce unnecessary experiments, and accelerate the identification and optimization of promising drug candidates.