ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Computational techniques for prediction of effects of small molecules on model organisms

Črtomir Gorup (2009) Computational techniques for prediction of effects of small molecules on model organisms. EngD thesis.

[img] PDF
Download (2005Kb)


    Studying the response of the model organisms exposed to chemicals can help us understand chemical activities, underlying biological processes and cell mechanisms. In the following disertation we have designed a set of computational techniques for predicting the effect of chemical compounds on model organisms. For chemical descriptors we have examined three different annotation systems, including QSAR based descriptors, molecular fingerprints (presence of specific short fragments) and MeSH terms from the MeSH ontology. Use of MeSH terms is also the the distinctive feature of our approach. We have developed a technique for computing MeSH term enrichment which enabled us to identify enriched subsets of chemicals with statistical significant ratio of chemicals with the target effect on phenotype. In order to identify the most suitable chemical description we have also developed a method for evaluating different types of attribute-based chemical descriptions. We used the support vector machine for predicting the effect of chemical compounds. Using the developed methods we analyzed the data from the experiment where model organism D. discoideum was exposed to 1.045 different chemical compounds and relative growth inhibition was observed as a phenotype. In general we are not able to predict the effects. However, if we split the chemicals to groups sharing some MeSH annotation term, we were able to find the terms for which our predictive procedures worked well. Results from the chemical description evaluation show that attributes based on MeSH terms are more suitable than QSAR-based descriptors and molecular fingerprints. We have also indentified 27 enriched (p < 0.02) MeSH terms which determine the same number of subsets with statisticaly significant ratio of chemical compounds causing the observing phenotype on model organism. Results confirm that use of MeSH terms improves prediction of the chemical impact on model organism.

    Item Type: Thesis (EngD thesis)
    Keywords: bioinformatics chemoinformatics machine learning data mining chemical structure analysis phenotypes MeSH ontology
    Number of Pages: 52
    Language of Content: Slovenian
    Mentor / Comentors:
    Name and SurnameIDFunction
    izr. prof. dr. Blaž Zupan106Mentor
    Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=50070&select=(ID=7290196)
    Institution: University of Ljubljana
    Department: Faculty of Computer and Information Science
    Item ID: 910
    Date Deposited: 14 Sep 2009 15:38
    Last Modified: 13 Aug 2011 00:35
    URI: http://eprints.fri.uni-lj.si/id/eprint/910

    Actions (login required)

    View Item