Projects / Programmes
Estimation of molecular properties with quantum chemical calculations and models
Code |
Science |
Field |
Subfield |
1.04.04 |
Natural sciences and mathematics |
Chemistry |
Organic chemistry |
Code |
Science |
Field |
P305 |
Natural sciences and mathematics |
Environmental chemistry |
P410 |
Natural sciences and mathematics |
Theoretical chemistry, quantum chemistry |
toxicity, carcinogenicity, 3d molecular structures, structure representation, quantum chemical calculations, modelling, QSAR
Researchers (3)
Organisations (1)
no. |
Code |
Research organisation |
City |
Registration number |
No. of publicationsNo. of publications |
1. |
0104 |
National Institute of Chemistry |
Ljubljana |
5051592000 |
21,007 |
Abstract
Due to the human activity many of new chemical compounds enter our environment. Some of the compounds are toxic or carcinogenic. Our interest is to collect the informations on their potential hazard (toxicity or carcinogenicity). The experimental evaluation of toxicity and carcinogenicity is expensive and time consuming task (over 1.5 mio USD per one compound, evaluations take over 2 years). The experiments are usually made on living things including mammals. Today we have confident data on carcinogenicity for only about 1200 compounds. Therefore, a big effort is dedicated to the developing of different computational methods for estimation of parameters describing the toxic or carcinogenic character of compounds. One of the many computational approaches is ''molecular structure - property'' modelling. Three questions are of crucial importance: (1) the representation of molecular structures, (2) the choice of the modelling method, (3) the selection of compounds for training of models. Ad (1): The geometrical structure of a molecule (3D structure) is determinated with the positions and atomic numbers of all atoms in a molecule. We developed a structural code, which enables a uniform representation of molecular structures (''''spectrum-like representation''''). Ad (2): Mathematical methods of modelling can be divided into two groups, linear and non-linear methods. We have been using linear regression method and counter propagation neural networks. Ad (3): The data of carcinogenicity are compiled by U. S. National Toxicology Program, governed by National Institute of Environmental Health Sciences. Furthermore, the models are tested on recall ability and with cross validation method. The goal of this project is to generate and test some ''molecular structure - toxicity/carcinogenicity'' models.