SCS Cancer Research Fellowships Awarded to Yutong Qiu and Trevor Frisby

The Carnegie Mellon School of Computer Science awarded one-year Cancer Research Fellowships to Yutong Qiu and Trevor Frisby in support of work toward their Ph.D. dissertations in the area of computational cancer biology. The awards are intended to support work that contributes to a vibrant community of cancer researchers at Carnegie Mellon, particularly emphasizing the university’s strength in computational and computationally-assisted directions in cancer research. Ms. Qiu and Mr. Frisby were selected for these awards based on compelling track records and research proposals describing how the awards would facilitate important directions they are pursuing in bringing computational methods to advancing cancer research and treatment.

Ms. Qiu was selected for a project entitled “Enabling Heterogeneity-aware Molecular Subtyping in Cancer.” The work will contribute to the topic of cancer molecular subtyping, which classifies cancer samples based on their molecular properties and has resulted in more precise diagnosis and treatment of cancer. Current cancer subtyping methods face two major challenges that compromise the accuracy of subtyping. First, the scope of these methods is limited to the gene expression landscape instead of covering the entire genome. Second, the current methods are not designed to capture the effect of sub-populations of genotypes, resulting from intra-tumor heterogeneity, on the biological properties of tumor samples. Ms. Qiu’s research aims to resolve these challenges by 1) enabling heterogeneity-aware sample comparison using genome graphs, and 2) capturing the molecular landscape of tumor samples using a combination of genomic variants and expression profiles. The central idea of her future work is to use a computational model known as a genome graph to compute similarities between mixtures of genotypes found in cancer. Her work will design measures to quantify tumor similarities based on the subgraphs that represent each genotype. This measure will not only account for the differences in genomic sequences but also the composition of genotypes within each sample. By integrating the similarity measure with gene expression profiles, this approach will comprehensively capture the molecular landscape of tumor samples and enable better cancer subtype identification – with potential applications to improving cancer diagnostics, better assigning patient-specific precision therapies, and identifying potential new targets for drug development. She will be advised in this work by SCS Computational Biology Department faculty member Carl Kingsford.

Mr. Frisby was selected for a project entitled “Generative Models to support the design of Cancer Vaccines & Immunotherapies.” His work will contribute to the area of cancer vaccines and immunotherapies which are among the most promising techniques for the prevention and treatment of cancer. The repertoire of immunotherapeutic strategies is diverse, owing to an extensive body of basic and clinical research, but many fundamental challenges remain, often creating barriers to clinical progress.  Mr. Frisby’s project seeks to address these challenges through the application of artificial intelligence (AI) and machine learning. Specifically, his work will develop a class of computational models known as a deep generative model and an algorithmic technique for working with such models known as Bayesian Optimization for application to several problems in cancer immunotherapeutics. Generative models are well suited to modeling complex evolutionary landscapes and their response to selective pressures, and the challenges of ongoing clonal evolution in cancers is one of the major reasons tumors often fail to respond to therapies or subsequently develop resistance to them. Specifically, the work will develop methods relevant to: (i) optimizing the design of antibody libraries against known tumor antigens; (ii) designing neoantigen-based vaccines; and (iii) identifying immune checkpoints. Mr. Frisby will be advised in this work by SCS Computational Biology Department faculty member Christopher James Langmead