Loading...
Analysis of DNA Methylation in Single-cell Resolution Using Algorithmic Methods and Deep Neural Networks
Rasti Ghamsari, Ozra | 2021
552
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 55003 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Sharifi Zarchi, Ali
- Abstract:
- DNA methylation in one of the most important epigenetic variations, which causes significant variations in gene expressions of mammalians. Our current knowledge about DNA methylation is based on measurments from samples of bulk data which cause ambiguity in intracellular differences and analysis of rare cell samples. For this reason, the ability to measure DNA methylation in single-cells has the potential to play an important role in understanding many biological processes including embryonic developement, disease progression including cancer, aging, chromosome instability, X chromosome inactivation, cell differentiation and genes regulation. Recent technological advances have enabled profiling DNA methylation at single-cell resolution. But, current protocols are limited due to the low coverage of CpGs and therefore, methods for predicting missing methylation states are essential for wider genome analysis. High-throughput sequencing is expensive and impractical in some fields and does not examine some parts of genome. Therefore, in this project, we use new supervised deep neural networks to predict methylation states in single-cells. For this purpose, methylation features of each CpG position and the DNA sequence patterns around them are used. The main purpose of project is extracting the related information from DNA sequence with CpG methylations using state of the art models, which can be applied on sequence data. Therefore, considering DeepCpG as the base model, the mentioned models are substituded as modules. Evaluations are done based on predicting methylation states and lagorithm’s performace, in terms of accuracy, precision, recall, f1-score and AUC-ROC metrics using real data from mouse embrionyc stem cells, and then comparing them with DeepCpG results. Results from testing the models on parts of data show that the new models, performed better on predicting methylated positions, but a little worse on unmethylated positions
- Keywords:
- Machine Learning ; Neural Network ; Single Cell RNA Sequencing (scRNA-seq) ; Gene Expression Data ; Multicellular Aggregates ; Epigenetic ; DNA Methylation