Combination of automated high throughput platforms, flow cytometry, and hierarchical clustering to detect cell state

Abstract

Background: This study examined whether hierarchical clustering could be used to detect cell states induced by treatment combinations that were generated through automation and high-throughput (HT) technology. Data-mining techniques were used to analyze the large experimental data sets to determine whether nonlinear, non-obvious responses could be extracted from the data. Methods: Unary, binary, and ternary combinations of pharmacological factors (examples of stimuli) were used to induce differentiation of HL-60 cells using a HT automated approach. Cell profiles were analyzed by incorporating hierarchical clustering methods on data collected by flow cytometry. Data-mining techniques were used to explore the combinatorial space for nonlinear, unexpected events. Additional small-scale, follow-up experiments were performed on cellular profiles of interest. Results: Multiple, distinct cellular profiles were detected using hierarchical clustering of expressed cell-surface antigens. Data-mining of this large, complex data set

Computationally efficient multidimensional analysis of complex flow cytometry data using second order polynomial histograms

“…Experience suggests that there is probably a general effect that adding more markers will clarify subsets that obscure each other in lower dimensions. Finally, increasing the number of parameters increases efficiency in screening for lymphoproliferative disorders in blood samples (42,43), or in high throughput screening for drug activity and off-target effects (20).…”

“…The most common approaches have been based on well-established kmeans clustering (18), but this requires an initial input of the number of clusters and is very sensitive to these initial conditions, as well as needing modifications to allow for nonGaussian data and nonspherical cluster shapes (16). Hierarchical clustering techniques (6,(19)(20)(21)(22), neighborhood maps based on Kullback-Leibler divergences (23) and principal component analysis (24) appear to be more suited to comparison between samples, and for patient classification schemes, than to unsupervised subset analysis within individual data files. Other methods have successfully used mixture models, but again require cluster number input, modification for non-Gaussian data, and have a very high computation load (25,26).…”

A probabilistic approach for the evaluation of minimal residual disease by multiparameter flow cytometry in leukemic B‐cell chronic lymphoproliferative disorders

“…In contrast, analysis of the data recorded has not attained the same level of progress and it is still based on strategies, which were defined more than 20 years ago (4,18,19). Accordingly, with a few exceptions (20)(21)(22)(23)(24)(25)(26) analysis of flow cytometry immunophenotypic data typically relies on the definition of a variable number of bidimensional plots, where an experienced operator selects the subpopulations of interest (25)(26)(27). Often, depending on the expertise of the operator, specific cell populations-particularly those present at low frequencies-can be misidentified.…”

Cytomics and nanobioengineering

“…On the other hand, it is the (mathematical) top-down approach to systems biology. Top-down approach means that based on the massive amount of data obtained by the high-content single cell measurements hypothesis-free data from diseased vs. healthy, treated vs. untreated or responders vs. nonresponders (patients) are compared to find the most informative data patterns for discriminating both groups (24,20) (example in Fig. 1).…”