Predviđanje protein-protein interakcija iz primarne strukture putem "Random Forest" klasifikatora

master's thesis

• Title (english): Prediction of protein-protein interactions from primary structure using a Random Forest classifier
• Author: Vedran Franke
• Mentor: Kristian Vlahoviček (mentor)
• Committee member: Kristian Vlahoviček (predsjednik povjerenstva)
• Committee member: Željka Vidaković Cifrek (član povjerenstva)
• Committee member: Ita Gruić-Sovulj (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Biology); Zagreb
• Defense date and country: 2010, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Biology

Abstract

Međusobne interakcije između proteina temelj su niza bioloških procesa, od regulacije metaboličkih puteva, specifičnosti imunoloških reakcija, replikacije DNK do sinteze proteina. Nagli razvoj visokoprotočnih metoda doveo je do velikog povećanja produkcije bioloških sekvenci, stvorivši potrebu za razvojem metoda i alata za njihovu funkcijsku analizu, te predviđanje fenotipskih svojstava, kako na molekularnoj, tako i na razini cijelog organizma. U ovom radu smo agregirali strukturalne podatke iz postojećih baza podataka, čime smo dobili skup proteinskih kvaternih struktura visoke kvalitete koji nam je omogućio primjenu metoda strojnog učenja za predviđanje interakcija između proteina. Iskoristili smo „Random Forest“ algoritam za predviđanje interakcijskih aminokiselina iz primarnih struktura proteina. Pokazali smo da, iako „Random Forest“ alogritam ima mogućnost klasifikacije visokodimenzionalnih podataka s izuzetnom točnošću, trenutno znanje o strukturalnim faktorima koji utječu na specifičnost interakcija između proteina nije na razini koja bi omogućila predviđanje interakcija na razlučivosti pojedinih aminokiselina koristeći isključivo sekvence proteina.

Abstract (english)

The interaction between proteins is fundamental to a broad spectrum of biological functions, including regulation of metabolic pathways, immunological recognition, DNA replication, progression through the cell cycle, and protein synthesis. Due to the growing disparity between the amount of sequenced genomic content and functional data, there exist a pressing need for tools and methods that will enable prediction of phenotypic traits, on the molecular or organism level, based on the sequence alone. In this work we have constructed a high quality dataset of protein structures that has enabled us to use the Random Forest non-linear classificator to develop a method for prediction of interacting residues from the protein primary structure. Our results have shown that, although the Random Forest algorithm has a unique capability of accurately classifying highly dimensional data, we still have an incomplete knowledge of structural factors that determine the specificity of protein-protein interactions, thus putting an upper limit the on the usefulness of the machine learning approach in predicting protein interactions on the level of single amino-acids.

• Keywords: interakcije proteina; strojno učenje; random forest; predikcija
• Keywords (english): protein interactions; random forest; machine learning; prediction
• Language: croatian
• URN:NBN: urn:nbn:hr:217:365613
• Study programme: Title: Biology; specializations in: Molecular biology; Course: Molecular biology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra biologije (magistar/magistra biologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2018-03-23 10:30:55