Traženje proteinskih motiva i klasifikacija

master's thesis

• Title (english): Search for protein motives and classification
• Author: Marko Iveković
• Mentor: Pavle Goldstein (mentor)
• Committee member: Pavle Goldstein (predsjednik povjerenstva)
• Committee member: Vanja Wagner (član povjerenstva)
• Committee member: Miljenko Marušić (član povjerenstva)
• Committee member: Marko Vrdoljak (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2022-05-05, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

U ovom diplomskom radu promatra se problem klasifikacije niza proteina određenog proteoma u neku proteinsku familiju. U svakom proteinu tražimo karakterističan podniz aminokiselina promatrane proteinske familije (motiv) te se tako problem svodi na klasifikaciju nizova aminokiselina na to jesu li oni motivi ili nisu. Koristeći postojeći iterativni algoritam dobiva se rezultat koji je skup potencijalnih motiva u kojem se nalazi popriličan broj lažno pozitivnih elemenata te je cilj eliminirati lažno pozitivne, a da se pritom sačuvaju oni koji su stvarno pozitivni, odnosno motivi. Rješavanju tog problema pristupa se tako da se promatra opis aminokiselina numeričkim faktorima te se dane nizove aminokiselina smješta u vektorski prostor, gdje se medu njima pokušava uočiti neka geometrijska struktura. Prethodno je razvijen algoritam koji u tom prostoru pronalazi središte kugle koja klasificira promatrane nizove aminokiselina, dok je pitanje radijusa ostalo otvoreno. U ovom radu se uz prethodno spomenuto smještanje nizova aminokiselina u vektorski prostor pronalazi procjena radijusa spomenute kugle. Izračunata procjena radijusa i središta kugle testira se na primjerima proteoma rajčice, krumpira i talijinog uročnjaka uz F1-score kao mjeru uspješnosti algoritma. Rezultati pokazuju da algoritam kojim se pronalazi središte i radijus odreduje kuglu koja je bila i cilj rada, a to je rješenje problema eliminiranja lažno pozitivnih nizova, a da se pritom sačuvaju oni koji su stvarno pozitivni. Uz to je taj algoritam brz te robustan na promjene veličine uzorka.

Abstract (english)

This thesis covers the classification problem of protein sequences of specified proteome into some protein family. In every protein we are searching for a characteristic subsequence of aminoacids (a motif) so the problem is reduced to classification of aminoacids. Using the existing iterative algorithm we get a set of potential motifs in which there is a certain number of false positive elements. Our goal is to eliminate the false positives while the true positive elements (motifs) remain. We approach the problem by examining a description of aminoacids in terms of numerical factors and by embedding the aminoacid sequences into a vector space and studying geometry of this embedding. An existing algorithm finds a center of a sphere which classifies the sequences, while the question of the radius remained open. In this thesis we estimate the radius of this sphere. The classification procedure is tested on three plant proteomes, using F1-score as a measure of accuracy. The results show that the algorithm gives a good solution to the problem. Furthermore, it is fast and robust with respect to the changes in data size.

• Keywords: proteom; proteinska familija; klasifikacija nizova aminokiselina; iterativni algoritam; F1-score
• Keywords (english): proteome; protein family; classification of aminoacids; iterative algorithm; F1-score
• Language: croatian
• URN:NBN: urn:nbn:hr:217:242086
• Study programme: Title: Mathematical Statistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2022-06-07 10:28:45