Primjena strojnog učenja u analizi i procjeni toksičnosti kemijskih spojeva

undergraduate thesis

• Title (english): Application of machine learning in the analysis and assessment of the toxicity of chemical compounds
• Author: Luka Buljeta
• Mentor: Mihaela Vranić (mentor)
• Committee member: Mihaela Vranić (predsjednik povjerenstva)
• Committee member: Damir Pintar (član povjerenstva)
• Committee member: Luka Humski (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-02, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad istražuje primjenu različitih modela strojnog učenja za binarnu klasifikaciju kemijskih spojeva u kontekstu njihove interakcije s određenim receptorima, koristeći Tox21 skup podataka. Istražena je učinkovitost triju modela: logistička regresija, slučajna šuma i XGBoost.
Logistička regresija procjenjuje vjerojatnost binarnih ishoda pomoću logističke funkcije. Model slučajne šume koristi mnoga stabla odlučivanja i kombinira njihove predikcije većinskim glasanjem. XGBoost iterativno trenira modele kako bi ispravio pogreške prethodnih, minimizirajući funkciju gubitka.
Podaci su očišćeni, podijeljeni na trening i test skupove, te su modeli trenirani i testirani koristeći metrike točnosti, preciznosti, odziva i F1-score-a. Eksperimenti su provedeni na dvije podjele podataka: 95% za treniranje i 5% za testiranje, te 75% prema 25%.
Rezultati pokazuju da je XGBoost model postigao najviše točnosti za većinu receptora u oba eksperimentalna postava, dok je model slučajne šume također pokazao visoke performanse, ali s većom varijabilnošću. Logistička regresija je općenito imala najniže performanse.

Abstract (english)

This paper investigates the application of various machine learning models for binary classification of chemical compounds in the context of their interaction with specific receptors, using the Tox21 dataset. The effectiveness of three models was explored: logistic regression, random forest, and XGBoost.
Logistic regression estimates the probability of binary outcomes using the logistic function. The random forest model employs multiple decision trees and combines their predictions through majority voting. XGBoost iteratively trains models to correct the errors of previous ones, minimizing the loss function.
The data were cleaned and split into training and test sets, and the models were trained and tested using accuracy, precision, recall, and F1-score metrics. Experiments were conducted on two data splits: 95% for training and 5% for testing, and 75% for training and 25% for testing.
The results show that the XGBoost model achieved the highest accuracy for most receptors in both experimental setups, while the random forest model also demonstrated high performance but with greater variability. Logistic regression generally had the lowest performance.

• Keywords: strojno učenje; kemionformatika; toksikologija
• Keywords (english): machine learning; chemoinformatics; toxicology
• Language: croatian
• URN:NBN: urn:nbn:hr:168:636647
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 20:14:43