| Summary: | Flow cytometry (FCM) is a technique for measuring physical, chemical and
biological characteristics of individual cells. Recent advances in FCM have
provided researchers with the facility to improve their understanding of the
tremendously complex immune system. However, the technology is hampered
by current manual analysis methodologies. In this thesis, I developed
computational methods for the automated analysis of immune response
FCM data to address this bottleneck. I hypothesized that highly accurate
results could be obtained through learning from the patterns that a biology
expert applies when doing the analysis manually.
In FCM data analysis, it is often desirable to identify homogeneous subsets
of cells within a sample. Traditionally, this is done through manual
gating, a procedure that can be subjective and time-consuming. I developed
SamSPECTRAL, an automated spectral-based clustering algorithm to
identify FCM cell populations of any shape, size and distribution while addressing
the drawbacks of manual gating. A particularly signi cant achievement
of SamSPECTRAL was its successful performance in nding rare cell
populations. Similarly, in most FCM applications, it is required to match
similar cell populations between di erent FCM samples. I developed a novel
learning-based cluster matching method that incorporates domain expert
knowledge to nd the best matches of target populations among all clusters
generated by a clustering algorithm.
Immunophenotyping of immune cells and measuring cytokine responses
are two main components of immune response FCM data analysis. I combined
the SamSPECTRAL algorithm and cluster matching to perform automated
immunophenotyping. I also devised a method to measure cytokine
responses automatically. After developing computational methods
for each of the above analysis components separately, I organized them into
a semi-automated pipeline, so they all work together as a uni ed package.
My experiments on 216 FCM samples con rmed that my semi-automated
pipeline can reproduce manual analysis results highly accurately both for
immunophenotyping and measuring cytokine responses.
My other main contributions were correlation analysis of intracellular and secreted cytokines, and developing a formula called GiMFI to improve
measuring functional response of cytokine-producing cells using
ow cytometry
assay.
|
|---|
