Performance comparison of supervised classifiers for detecting leukemia cells in high-dimensional mass cytometry data

Advances in highly multi-parametric measurements by mass cytometry have made possible the accurate detection of acute myeloid leukemia (AML) cells in complex cell populations. However, current informatics methods bottlenecks data processing by being labor-intensive, time-consuming, and prone to user bias. To address these problems, major efforts have been made to automate the detection of AML cells. A key step for the task is to choose a proper supervised algorithm. This paper describes a comparative study on the performance of three algorithms: support vector machines, back propagation artificial neural network and naive Bayes. We considered practical situations involving few training samples and minor AML cells. Characteristic features, such as classification accuracy, running time, etc. were observed for asserting the performance of the three classifiers. Our experiments indicated that SVM produced superior results to the other two classifiers.