7.1 Introduction to MICA

MICA (Mutual Information-based Clustering Analysis) is a clustering tool designed for single cell genomics data. Compared to most existing single-cell clustering algorithms, MICA has two unique features:

• MICA uses mutual information to measure cell-cell similarity for unsupervised clustering analysis, while most existing tools employ linear-transformation of PCA and/or co-expression analysis using linear Pearson or Spearman correlations that may not capture the nonlinear cell-cell distance.
• MICA uses all high-quality features for clustering, while most existing tools select the top highly variable features to improve the clustering speed. This is arbitrary and may lose the information that can distinguish close cell states.

MICA is developed using Python framework, to take its strengths in calculation speed and memory consumption. A lot of effort has been made to improve the interoperability between Python and R. Now MICA works seamlessly with the SparseExpressionSet object. The input of MICA can be easily generated from the SparseExpressionSet object by generateMICAinput(), and the output of MICA, the clustering results, can be effortlessly visualized by MICAplot() and integrated into SparseExpressionSet object by addMICAoutput().