Date

2022-8-31

Author

Yaman, Oğuz Ulaş

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

196
views

75
downloads

Cite This

In this MSc thesis, the aim is to propose methodologies for predicting transcription factor binding sites in yeast cells. This aim is achieved, first, by modeling P. pastoris central carbon metabolism genes using phylogenetic footprinting; and next, by modelling S. cerevisiae transcription factors’ affinity towards 8-mers using the Machine Learning algorithmic models, i.e., Random Forest, XGBoost, and Deep Learning. In the first part of the thesis, a novel phylogenetic footprinting algorithm is introduced, which requires any number of orthologous promoter pairs with their DNA sequences, and a database that contains the transcription factor binding motifs as input. The model first scans the reference promoter for TF binding sites, and then using pairwise alignment, determines the conserved transcription factor binding sites in the target promoter. The algorithm was used to compare 58 S. cerevisiae promoters of the genes in the central carbon metabolism with the predicted 52 orthologous P. pastoris promoters. The presented phylogenetic footprinting predictions of transcription factor binding sites enabled annotation of 116 P. pastoris transcription factors in the central pathways. In the second part of the thesis, seven Machine Learning algorithmic models (five based on Neural Networks, one based on XGBoost, and one based on Random Forest) were trained to predict high affinity 8-mers for S. cerevisiae transcription factors. The 8-mers were represented embedded into numerical arrays with using the predetermined five features that can represent sequence specificities of the transcription factor binding sites. Since different transcription factors may recognize different features, A greedy approach was designed, which selectively picks the best pool of features and makes the model combination for each transcription factor that gives the best Matthews Correlation Coefficient (MCC) score on test data. The presented novel approach yielded an average MCC score of 0.873 in predicting high-affinity binding sites for all the transcription factors.

Subject Keywords

Transcription Factor, Transcription Factor Binding Site, Yeast, Phylogenetic Footprinting, Machine Learning

URI

https://hdl.handle.net/11511/98795

Collections

Graduate School of Natural and Applied Sciences, Thesis