Analiza točnosti pretraživanja

U ovom radu koristili smo tehniku analiziranja određenog skupa proteina pomoću najveće maksimalne klike kako bismo pronašli proteine iz iste proteinske familije. Analizirali smo točnost te pretrage, te smo usporedili dobivene rezultate s rezultatima dobivenim iterativnim pretraživanjem proteoma. Pomoću očekivanih parametara dobivenih linearnom regresijom na proteinima iz iste familije, izračunali smo očekivanu funkciju sličnosti i na taj način pridružili promatranom skupu proteina težinski, a zatim i \(0-1\) graf kojem smo odredili najveću maksimalnu kliku. Tehniku smo proveli na proteomima talijinog uročnjaka, rajčice, riže i šećerne repe. Dobiveni zaključak je da, uz optimalne parametre, metodom najveće maksimalne klike poboljšavamo točnost pretrage. Za ocjenu modela koristili smo \(\mathrm{F_1}\)-score, koji je u odnosu na iterativno pretraživanje uz izabrane parametre značajno porastao na svim odgovorima dobivenima najvećom maksimalnom klikom. Također, u kliki je ostalo sačuvano više od 97% stvarnih pozitivaca iz uzorka.In this thesis, the largest maximal clique analyzing method has been used on a specific collection of proteins to find proteins belonging to the same protein family. The accuracy of that search has been analyzed and the results have been compared with the results obtained by the iterative search of the proteome. Using expected parameters obtained by the linear regression on the proteins from the same protein family, the expected similarity function has been calculated and associated with the weight graph and \(0-1\) graph on the specific collection of proteins. Then the largest maximal clique for a specific collection of proteins in the \(0-1\) graph has been found. The method was used on proteomes of Arabidopsis thaliana, Solanum lycopersicum, Oryza sativa and Beta vulgaris. The conclusion is that the largest maximal clique method, with the optimal parameters, improves the accuracy of the search. The \(\mathrm{F_1}\)-score was used to assess the results, and the largest maximal clique model yielded a significant improvement compared to the plain iterative search method. Furthermore, the clique preserved more than 97% of true positives in the sample