Cell soup in screens

In a transcription factor screen, genes from many factors are implanted into cells to determine their effects, either singly or in combination. The stress of this process, as well as later sorting can cause cells to burst, spilling their contents into the surrounding media. As a result, sequenced cells can show contamination from other transcription factors, complicating efforts to label cells with their transfected factors. This paper develops a systematic approach to this problem by (1) deriving an exact form for the distribution of observed noise counts assuming that they are the result of cells bursting and mixing through a large volume, and (2) applying this distribution to label cells with their likely factors and compute associated uncertainties in this labelling. The model suggests that the noise in different genes is generated by a shared process and we show evidence for this in the results of a screening experiment. The labels and uncertainties can then be translated into better informed inferences on the effects of factors on cell type and function, which is essential in the low-signal, high-noise world of biological data.