Expectation-Maximization Based Mixture-Model Exploiting Pathway Knowledge for Cancer Heterogeneity
Abstract
Cancer cells are much more prone to mutations than normal cells, generating over time, more genetic variants of themselves within the tissue. Drugs designed for one variant might not work as intended for other variants. As such, effective drug design requires estimation of proportion of various cancer subpopulations.
In this work, a mixture model based approach with expectation maximization is proposed for determination of cancer heterogeneity. We exploit the pathway knowledge collected by biologists over time to surpass the limitations of identifiability shown by mixture models. Also in cases where Expectation-Maximization converges to more than one solution, pathway knowledge is used to break the tie by defining an error metric. Finally, using experimental data, changes in composition of the mixture over time are estimated using the model. The approach can also be used to compare the effectiveness of different drugs on a heterogeneous cancer tissue by observing the response over time.
Citation
Kapoor, Rajan (2017). Expectation-Maximization Based Mixture-Model Exploiting Pathway Knowledge for Cancer Heterogeneity. Master's thesis, Texas A & M University. Available electronically from https : / /hdl .handle .net /1969 .1 /161604.