Fig 1. Representative workflow for the enhanced image-based cell profiling via data cleaning.

Six main steps transform cell measurements of each plate to compound-level profiles. Specifically, in step 3, we replace each feature by part of those features that are accounted for by the cell area feature. This is achieved by regressing each feature against the cell area, and replacing each feature by the residue of the corresponding regression. Furthermore, cells that appear as outliers are identified and removed through a histogram-based approach, which is called HBOS, to make sure that the well-level aggregated profiles would not be affected by potential artifacts.