Predviđanje mikrobioma tla metodama strojnog učenja

master's thesis

• Title (english): Machine learning methods for soil microbiome prediction
• Author: Marko Šandrk
• Mentor: Mirjana Domazet-Lošo (mentor)
• Committee member: Mirjana Domazet-Lošo (predsjednik povjerenstva)
• Committee member: Ivica Botički (član povjerenstva)
• Committee member: Tomislav Jagušt (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-02-23, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Razvojem novih metoda visokopropusnog sekvenciranja dolazi se do sve većih
količina genomskih i metagenomskih podataka. Za analizu tih podataka sve se češće
koriste i algoritmi strojnog i dubokog učenja.
Mikrobiom, zajednica mikroorganizama koja obitava u nekoj okolini jest ključni
stup života na Zemlji te bolje razumijevanje ovih zajednica može imati pozitivne učinke
od ljudskog zdravlja do ekonomije.
U sklopu rada razvijen je niz modeli dubokog učenja bazirani na nepotpunom te
regulariziranom autoenkoderu za predviđanje mikrobioma tla na temelju različitih značajki okoliša kao što su temperatura, oborine, prisutna kultura i slično. Posebno se dobrim
pokazala uporaba regulariziranih autoenkodera koji su postigli čak i nešto bolje
rezultate od modela razvijenih u studiji (García-Jiménez et al., 2020).

Abstract (english)

With the development of new high throughput sequencing methods more and more
genomic and metagenomic data is gathered. Machine and deep learning methods are
used for analysis of such data more frequently.
Microbiome, a community of microorganisms cohabitating in some dedicated environment,
is a vital part of life on the planet Earth. Better understanding of these communities
could have positive effects on human health as well as economy.
In this thesis, several deep learning models are developed for purposes of predicting
soil microbiome based on a few environmental features such as temperature, precipitation
etc. All trained models were based on an undercomplete or a regularized autoencoder
architecture. Using regularized autoencoders proved particularly convenient for
this problem achieving slightly better scores on used metrics than those provided by
(García-Jiménez et al., 2020).

• Keywords: duboko učenje; nepotpuni autoenkoder; regularizirani autoenkoder; OTU; OTU tablica
• Keywords (english): deep learning; undercomplete autoencoder; regularized autoencoder; OTU; OTU table
• Language: croatian
• URN:NBN: urn:nbn:hr:168:891141
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 16:13:58