Integration and analysis of scRNAseq data using genome-scale metabolic models

Nastran, Jurij (author), Moškon, Miha (mentor), Mraz, Miha (comentor)

Recent advancements in high-throughput biological screening have led to the emergence of large datasets of single-cell RNA sequencing data. These new datasets enable us to better understand the underlying biological systems behind diseases such as prostate cancer. We introduce a linear discriminant analysis-based method to effectively filter out the least informative genes from the obtained single-cell RNA dataset containing prostate cancer cells, thereby enhancing the dataset's discriminatory power. We also develop a new non-parametric genome-scale metabolic model extraction method. We then use this method as well as a publicly available genome-scale metabolic model to generate reaction flux data of each cell. We compare the reaction flux data with the single-cell RNA data by training and analyzing separate random forest models to classify cell types from samples with prostate cancer. The improved classification results as well as the analysis of attribute importances show the single-cell RNA sequencing data to be more informative and thus better suited for the task of effectively differentiating the cell types from samples containing prostate cancer.

• Language: English
• Keywords: genome-scale metabolic models, model extraction methods, context-specific metabolic models, omics data integration, metabolism, cancer, single-cell RNA sequencing data
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-155937
• COBISS.SI-ID: 195566595
• Publication date in RUL: 24.04.2024

Secondary language

• Language: Slovenian
• Title: Integracija in analiza podatkov scRNAseq z uporabo metabolnih modelov na nivoju genoma

Abstract

Nedavni napredek v visoko zmogljivem sekvenciranju bioloških podatkov je privedel do nastanka velikih podatkovnih naborov RNK na ravni posameznih celic. Ti novi nabori podatkov nam omogočajo boljše razumevanje bioloških sistemov, ki so v ozadju bolezni, kot je rak prostate. Predstavljamo metodo, ki temelji na linearni diskriminantni analizi, s katero učinkovito filtriramo najmanj informativne gene iz pridobljenih podatkov o RNK na ravni posameznih celic, ki vsebujejo celice raka prostate. S to metodo izboljšamo razločevalno moč nabora podatkov. Razvijemo tudi novo metodo za ekstrakcijo modelov brez potrebe po ročni določitvi parametrov. Nato uporabimo to metodo skupaj z javno dostopnim metabolnim modelom na nivoju genoma za generiranje podatkov o reakcijskem toku znotraj vsake celice. Nato te podatke primerjamo s podatki o RNK na ravni posameznih celic in analiziramo ločene modele naključnih gozdov za klasifikacijo celic iz vzorcev raka prostate. Rezultati klasifikacije in analiza pomembnosti atributov kažejo, da so podatki o RNK bolj informativni za razlikovanje vrst celic iz vzorcev raka prostate kot podatki o reakcijskem toku.

• Keywords: metabolični modeli na nivoju genoma, metode ekstrakcije modelov, kontekstno-specifični metabolični modeli, integracija omičnih podatkov, metabolizem, tumor, sekvenčni podatki RNA na nivoju celice