Novel machine learning methods for cancer sequencing analysis
<p>Heterogeneity is arguably one of the most important hallmarks of cancer which contributes to its drug resistance property. Cancer heterogeneity is the consequence of an evolutionary process during its development. Understanding cancer evolution will thus benefit the drug development field a...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: |
| Summary: | <p>Heterogeneity is arguably one of the most important hallmarks of cancer which contributes to its drug resistance property. Cancer heterogeneity is the consequence of an evolutionary process during its development. Understanding cancer evolution will thus benefit the drug development field as well as clinical treatment of cancer.</p>
<p>In this thesis, I developed three novel approaches based on machine learning for the analysis of \correction{cancer evolution} using genomics data.</p>
<p>In the work related to mutation signature, I developed a novel approach using convolution filtering to detect new signatures. I showed that this new approach works better than existing method under certain condition and could be applied to longer nucleotide sequences.</p>
<p>In the work for copy number evolution, I developed a reinforcement learning approach to uncover the evolution history. I showed that this approach could recover the history more accurately than existing methods. It is also the first reinforcement learning approach developed in the field for cancer evolution.</p>
<p>In the work for high dimensional genomic data analysis, I developed an approach which jointly carries out dimension reduction and clustering. I showed that this method could recover the underlying latent structure of genomic sequencing data more accurately than existing methods and this method could be used for scRNA-seq data analysis as well as many other fields.</p>
<p>These novel methods developed could be used for more cancer genomic data to gain better understanding of the cancer evolution process in the future.</p> |
|---|