A0700
Title: Robust clustering using maximized mutual information
Authors: Mackenzie Neal - McMaster University (Canada) [presenting]
Paul McNicholas - McMaster University (Canada)
Arthur White - Trinity College Dublin (Ireland)
Abstract: Flow cytometry technology allows for the analysis of disease impact on individual cells. Population identification in flow cytometry datasets is commonly performed, with manual gating being the most frequently used method. However, manual gating is subjective, often irreproducible, and becoming increasingly more difficult to perform as the complexity of flow cytometry experiments and datasets grows. Methods have been proposed to automate population identification of flow cytometry data; many such methods rely on generative clustering models. Ideas from both generative and discriminative models are incorporated to present a clustering algorithm capable of capturing irregular sub-populations common in biological datasets. Although flow cytometry is the primary motivation for the work proposed herein, this method can be applied to any dataset wherein the goal is to obtain intuitive clustering solutions. This is demonstrated by comparison to popular clustering methods on various simulated and real datasets.
