APPLICATION OF GENETIC FUNCTION APPROXIMATION TO QUALITATIVE STRUCTURE-ACTIVITY RELATIONSHIPS

QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIP (QSAR) ANALYSIS IS THE AREA OF RESEARCH WHICH TRIES TO BUILD SUCH COMPUTATIONAL MODELS OF BIOLOGICAL ACTIVITY USING THE PHYSICOCHEMICAL PROPERTIES OF A COMPOUND. THE BASIC ASSUMPTION IS THAT THE VARIATIONS OF BIOLOGICAL ACTIVITY WITHIN A SERIES OF SIMILAR STRUCTURES CAN BE CORRELATED WITH CHANGES IN EXPERIMENTAL OR COMPUTATIONAL PARAMETERS OF THE MOLECULES. THESE PARAMETERS COULD MEASURE, FOR EXAMPLE, HYDROPHOBIC, STERIC, AND ELECTRONIC PROPERTIES WHICH MAY INFLUENCE BIOLOGICAL ACTIVITY. IN THIS PROCESS, A DATA TABLE IS FORMED WHERE EACH ROW REPRESENTS A CANDIDATE COMPOUND, AND EACH COLUMN A COMPUTATIONAL OR EXPERIMENTAL PROPERTY. STANDARD REGRESSION ANALYSIS IS APPLIED TO THIS DATA TABLE TO CREATE A MODEL OF ACTIVITY BASED ON THE PARAMETERS. THE GENETIC FUNCTION APPROXIMATION (GFA) ALGORITHM OFFERS A NEW APPROACH TO THIS PROBLEM. RESEARCHERS ARE REPLACING STANDARD REGRESSION ANALYSIS WITH THE GFA ALGORITHM IN THE QSAR TASK. THIS ALLOWS THE CONSTRUCTION OF HIGHER-QUALITY PREDICTIVE MODELS, EVEN WITH DATA SETS THAT MAY HAVE THOUSANDS OF PARAMETERS. INITIAL STUDIES USING GFA SHOW THE ALGORITHM IS CAPABLE OF BUILDING MODELS WHICH MEET OR EXCEED THE PREDICTIVE ABILITY OF THE BEST PUBLISHED MODELS. THE ALGORITHM IS ABLE TO OUTPERFORM NOT ONLY STANDARD REGRESSION ANALYSIS IN THIS AREA, BUT ALSO NEWER ALGORITHMS BASED ON PLS OR NEURAL-NETWORKS. RAPID COMMERCIALIZATION OF THIS TECHNOLOGY WOULD LEAD TO A POWERFUL NEW QSAR TOOL, AND COULD HAVE SIGNIFICANT BENEFIT IN THE ABILITY OF QSAR TOOLS TO AID CHEMISTS IN DRUG DESIGN, AND THUS REDUCE THE CURRENT DEPENDANCE ON EXCESSIVE TESTING DURING DRUG DEVELOPMENT.