Computational Frameworks to Interpret Genetic Abnormalities in Hematologic Malignancies

In the era of high throughput sequencing technologies, the speed of genomic data generation exceeds our abilities to store, analyze and interpret the data without high-performance computing facilities. In hematology, next-generation sequencing (NGS) is being utilized for diagnosis, prognostication, and post-treatment monitoring. Although it is undoubtedly the future of molecular testing and monitoring, computational methods to interpret results at single and longitudinal time points are highly in need. Thus, this thesis is devoted to mining clinically and scientifically meaningful information from NGS data in the context of hematologic malignancies. First, I show how incorporating allelic burdens of mutations can enhance the risk stratification of patients with acute myeloid leukemia (AML). Instead of considering mutations as binary variables, the model considers mutations as continuous variables, considering clonal architecture when predicting the prognosis. Also, I demonstrate single mutation does not determine a patient's risk alone, but the patient's risk depends on the entire mutation profile. Second, I describe computational methods to detect residual leukemia cells remaining post-treatment (termed "measurable residual disease" (MRD)). I show that NGS-based MRD detection is feasible in different subtypes of AML at various time points. In addition, estimating sequencing error rates helps reduce false positives while increasing true positives, thus maximizing clinical relevance. Throughout multiple studies, I describe different approaches to detect NGS-based MRD and to mine clinically meaningful information. Lastly, I present a computational workflow that analyzes single-cell proteogenomic sequencing data to identify relapse-fated subclones from the time of diagnosis in an FLT3-ITD+ AML case. Simultaneous proteogenomic profiles allow the refinement of clonal models relying only on mutation profiles and help identify relapse-fated AML subclones at the time of initial diagnosis. I also describe how copy-neutral loss-of-heterozygosity at chr13q can be accurately detected at the single-cell level and cell surface marker expressions help identify somatic mutations.

Department dc:contributor.department

Computer Science

Year dc:date.issued

2023

Author dc:creator

• Kim, Taehyung

Advisor dc:contributor.advisor

• Zhang, Zhaolei ZZ

• Acute myeloid leukemia
• Cancer genomics
• Computational biology
• Hematologic malignancies
• Machine learning
• Measurable residual disease

Handle dc:identifier.uri

http://hdl.handle.net/1807/128064

OAI identifier oai:identifier

oai:utoronto.scholaris.ca:1807/128064