Richard Scheuermann & Tim Mosmann: Automatic flow cytometry gating methods (2012 CBIM Summer School)

Class session June 14 2012, 9:00-10:15AM including discussion Introduction to Flow Cytometry data - why is manual processing no longer adequate? Large datasets Multiple parameters, up to 20 or even more Diverse cell populations Compensation, normalization, standardization Current practice depends on subjective manual gating - each person different Main problem - humans don't visualize in more than three dimensions. Data pre-processing Normalization, different reagents, machine settings, flow cytometers, Compensation - current algorithms not accurate Compensation limits, background versus fluor compensation Transformations to handle negative numbers on a log scale - arcsinh, logicle etc. Clustering - goals Ideally, one cluster = one biological population In practice, actual diversity is probably flow cytometry diversity Clustering is a practical separation of useful populations Ambiguity - decision about when to split a population further Ambiguity - alternative solutions Clustering methods: Model-based, Gaussian, skewed Gaussian, etc Grid-based Thresholding, Watershed SPADE Others..... Sample comparison and inference Cluster matching - difficult, inter-related with normalization How to incorporate expert knowledge? Evaluation of data processing algorithms Difficulty of obtaining ground truth - how to define "correct" clustering? Evaluating cells versus evaluating populations