Methods for Image-Based Cell Profiling
Aggregate data based on given grouping.
Remove redundant variables.
Count the number of NAs per variable.
Compute covariance matrix and vectorize.
Remove variables with NA values.
Drop rows that are NA in all specified variables.
Extract subpopulations.
Generalized log transform data.
A sparse matrix for sparse random projection.
Normalize observation variables.
Measure replicate correlation of variables.
Reduce the dimensionality of a population using sparse random projecti...
Feature importance based on data entropy.
Transform observation variables.
Measure variable importance.
Select observation variables.
Remove variables with near-zero variance.
Whiten data.
Typical morphological profiling datasets have millions of cells and hundreds of features per cell. When working with this data, you must clean the data, normalize the features to make them comparable across experiments, transform the features, select features based on their quality, and aggregate the single-cell data, if needed. 'cytominer' makes these steps fast and easy. Methods used in practice in the field are discussed in Caicedo (2017) <doi:10.1038/nmeth.4397>. An overview of the field is presented in Caicedo (2016) <doi:10.1016/j.copbio.2016.04.003>.