Skip to Main content Skip to Navigation
New interface

Detection of structural variants in cancer genomes using a Bayesian approach. You will find below the abstract of my PhD thesis

Daria Iakovishina 1, 2 
2 AMIB - Algorithms and Models for Integrative Biology
LIX - Laboratoire d'informatique de l'École polytechnique [Palaiseau], LRI - Laboratoire de Recherche en Informatique, UP11 - Université Paris-Sud - Paris 11, Inria Saclay - Ile de France
Abstract : According to the current knowledge, cancer develops as a result of the mutational process of the genomic DNA. In addition to point mutations, cancer genomes often accumulate a significant number of chromosomal rearrangements also called structural variants (SVs). Some types of cancer are associated with recurrent SVs (e.g., ABL/BCR in chronic myelogenous leukemia, EWSR1/FLI1 in Ewing sarcoma, amplification of MYCN in neuroblastoma, amplification of ERBB2 in ovarian and breast cancers, SV in the promoter of TAL1 in lymphoblastic leukemia). It is important to be able to identify exact positions and types of these variants to be able to track cancer development or select the most appropriate treatment for the patient. Next generation sequencing (NGS) technologies provide a possibility to identify SVs in a very precise and time-efficient manner. In my PhD work, I propose a new computational method named SV-Bay, aimed to detect structural variants using whole genome sequencing data. This method combines two SV detection techniques into one: it takes into account both paired-end mapping abnormalities and variation of the depth of coverage. SV-Bay uses a probabilistic Bayesian approach to combine these techniques. SV-Bay statistical model includes possible sequencing errors, read mappability profile along the genome and changes in the GC-content. On one hand, our approach is capable to accurately filter out false candidate SV, and thus has an improved SV detection precision. On the other hand, the fact of taking into account read mappability and GC-content makes of SV-Bay an extremely sensitive approach. In addition, SV-Bay includes a possibility to keep only somatic SVs if matched normal control data are provided. I performed a comparison of SV-Bay with 4 widely used SV detection tools: Delly, GASVPro, Lumpy and BreakDancer (BreakDancerMax). For this comparison, I used simulated mate-pair and paired-end datasets along with real mate-pair data for the CLB-GA neuroblastoma cell line. For all simulated datasets, the proposed method showed the best results in terms of both recall and precision. For the experimental neuroblastoma dataset, structural variants discovered by SV-Bay explained 78% of breakpoints in the copy number profile, calculated using an Affymetrix SNP6.0 array, while providing a much smaller number of candidate SVs than the three other tools. As a part of my PhD work, I also constructed a novel exhaustive catalogue of SV types. Based on the previous publications and experimental data I introduced a list of 17 structural variant types; this list included seven SV types ignored by the existing SV calling algorithms. This, to date the most comprehensive SV classification, was used in the SV-Bay method to annotate predicted SVs.
Document type :
Complete list of metadata

Cited literature [52 references]  Display  Hide  Download
Contributor : Daria Iakovishina Connect in order to contact the contributor
Submitted on : Sunday, March 27, 2016 - 4:26:29 PM
Last modification on : Tuesday, October 25, 2022 - 4:18:28 PM
Long-term archiving on: : Monday, November 14, 2016 - 6:43:04 AM


  • HAL Id : tel-01294142, version 1


Daria Iakovishina. Detection of structural variants in cancer genomes using a Bayesian approach. You will find below the abstract of my PhD thesis. Computer Science [cs]. Ecole Polytechnique, 2015. English. ⟨NNT : ⟩. ⟨tel-01294142⟩



Record views


Files downloads