Using the systematic nature of errors in NGS data to efficiently detect mutations : computational methods and application to early cancer detection

Abstract : Comprehensive characterization of DNA variations can help to progress in multiple cancer genomics fields. Next Generation Sequencing (NGS) is currently the most efficient technique to determine a DNA sequence, due to low experiment cost and time compared to the traditional Sanger sequencing. Nevertheless, detection of mutations from NGS data is still a difficult problem, in particular for somatic mutations present in very low abundance like when trying to identify tumor subclonal mutations, tumor-derived mutations in cell free DNA, or somatic mutations from histological normal tissue. The main difficulty is to precisely distinguish between true mutations from sequencing artifacts as they reach similar levels. In this thesis we have studied the systematic nature of errors in NGS data to propose efficient methodologies in order to accurately identify mutations potentially in low proportion. In a first chapter, we describe needlestack, a new variant caller based on the modelling of systematic errors across multiple samples to extract candidate mutations. In a second chapter, we propose two post-calling variant filtering methods based on new summary statistics and on machine learning, with the aim of boosting the precision of mutation detection through the identification of non-systematic errors. Finally, in a last chapter we apply these approaches to develop cancer early detection biomarkers using circulating tumor DNA
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-02311750
Contributor : Abes Star <>
Submitted on : Friday, October 11, 2019 - 10:24:08 AM
Last modification on : Friday, October 11, 2019 - 10:24:10 AM

File

TH2019DELHOMMETIFFANY.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02311750, version 1

Collections

Citation

Tiffany Delhomme. Using the systematic nature of errors in NGS data to efficiently detect mutations : computational methods and application to early cancer detection. Bioinformatics [q-bio.QM]. Université de Lyon, 2019. English. ⟨NNT : 2019LYSE1098⟩. ⟨tel-02311750⟩

Share

Metrics

Record views

48

Files downloads

4