Probabilistic subclonal reconstruction for cancer

Kim, Juho

Permalink

https://hdl.handle.net/2142/116102 Copy

Description

Title

Probabilistic subclonal reconstruction for cancer

Author(s)

Kim, Juho

Issue Date

2022-07-14

Director of Research (if dissertation) or Advisor (if thesis)

• Koyejo, Oluwasanmi
• El-Kebir, Mohammed

Doctoral Committee Chair(s)

• Koyejo, Oluwasanmi
• El-Kebir, Mohammed

Committee Member(s)

• Milenkovic, Olgica
• Shomorony, Ilan
• Chia, Nicholas

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Cancer genomics
• Tumor phylogenetics, Probabilistic modeling
• Machine learning

Abstract

Cancer consists of genetically heterogeneous populations of cells that arise through a process of subclonal evolution. Reconstructing the evolutionary processes that give rise to cancer can help us better understand cancer progression and prioritize treatment targets. The subclonal reconstruction of cancer gives us the information about the co-occurrence of mutations within the same subclone, the underlying proportion of cells belonging to each subclone, and the ancestral relationships between them. The evolutionary process can be described by inferring tumor phylogenetic trees. The majority of current approaches focus only on either mutation clustering or tree inference in isolation, or rely on computationally expensive algorithms to holistically consider clustering and tree inference concurrently. In this dissertation, we formalize the problem of reconstructing subclonal structure for cancer via probabilistic modeling. Using variant and total read count obtained from bulk DNA sequencing data as input, we introduce a tree-constrained binomial mixture model and an expectation-maximization (EM) method to estimate the clustering assignment for each mutation and the underlying frequency for each cluster. Our EM algorithm employs a linear programming approach to accurately maximize the likelihood bound subject to tree constraints. We choose the optimal tree topology by repeating the process across all possible tree topologies. Compared to existing work, the resulting ClusTree algorithm more accurately identifies mutation clusters, estimates frequencies for each cluster, and detects the proper tree topology, especially for low-depth sequencing data.

Graduation Semester

2022-08

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Juho Kim

Owning Collections