1.

Computational method for modeling of gradient separation in ion-exchange chromatography

A model for the simulation of the gradient separation in ion-exchange chromatography is based on discontinuous plate model and simulates the distribution of analytes in the ion-exchange column during the separation process. It enables calculations of chromatograms for the analytes with integer and non-integer effective charges under complex gradient profiles. The observed average of the absolute values of the relative errors of the retention times obtained for any analyte in the present study is below 4%, while the average error considering all analytes in the study is below 2%.

COBISS.SI-ID: 4202266

2.

Bottle-neck type of neural network as a mapping device towards food specifications

A novel methodology is proposed for food specifications associated with the origin of food. The methodology was tested on honey samples collected within the TRACE EU project. The data were sampled in various regions in Europe and analysed for the trace elements content. The sampling sites were characterized by different geological origins, such as limestone, shale, or magmatic. The novel methodology proposed for food specifications was demonstrated on a reduced set of samples, which shows good clustering of all three classes, and on the third class of the original data set.

COBISS.SI-ID: 4194842

3.

Automatic adjustment of the relative importance of different input variables for optimization of counter-propagation artificial neural networks

We present a quantitative structure-activity relationship study with 49 peptidic molecules, inhibitors of the HIV-1 protease. The modelling was preformed using counter-propagation artificial neural networks (CPANN), which has been proven as a valuable tool for data analysis. With the genetic algorithm, the relative importance was adjusted during the training of CPANN. The proposed approach is capable of finding simpler efficient models in comparison with the original approach of equally important input variables. A simpler model is also more robust and less subjected to the overfitting model.

COBISS.SI-ID: 4088090

4.

Data-driven model for the prediction of protein transmembrane regions

The data about sequences and transmembrane regions were collected from PDB (Protein Data Bank). The entire studied set consists of 5800 segments. Protein sequences were represented with 20x20 matrices, where each element indicates a pair of neighbouring amino-acids. Using the couterpropagation neural network we construct the model which separates the transmembrane regions. The model was presented also at the conference CMTPI 2009 and the poster was awarded with the first prize.

COBISS.SI-ID: 4344090

5.

Kohonen and counter-propagation neural networks employed for modeling endocrine disruptors

The mehodology for modelling and classification of endocrine disruptors is based on Kohonen and counter propagation neural networks. Three data sets were considered: 106 substances extracted from the list of 553 of known endocrine disruption activity, 132 compounds tested for their binding affinities to the mice ER, and 60 chemicals tested for the binding affinity to human ER alpha and beta. To obtain the structure-activity relationship, the 3D structures of complexes were taken into account. Structural features of ligands with the strongest influence to the binding affinities were detected.

COBISS.SI-ID: 4087322