Genetic programmingforanticancertherapeuticresponsepredictionusingthe NCI-60 dataset

Statistical methods,andinparticularmachinelearning,havebeenincreasinglyusedinthedrugdevelop- ment workflow.Amongtheexistingmachinelearningmethods,wehavebeenspecificallyconcernedwith genetic programming.Wepresentageneticprogramming-basedframeworkforpredictinganticancer therapeutic response.WeusetheNCI-60microarraydatasetandwelookforarelationshipbetweengene expressions andresponsestooncologydrugsFluorouracil,Fludarabine,FloxuridineandCytarabine.We aim atidentifying,fromgenomicmeasurementsofbiopsies,thelikelihoodtodevelopdrugresistance. Experimental results,andtheircomparisonwiththeonesobtainedbyLinearRegressionandLeastSquare Regression, hintthatgeneticprogrammingisapromisingtechniqueforthiskindofapplication.More- over, geneticprogrammingoutputmaypotentiallyhighlightsomerelationsbetweengeneswhichcould support theidentificationofbiologicalmeaningfulpathways.Thestructuresthatappearmorefrequently in the“best”solutionsfoundbygeneticprogrammingarepresented.