CausalCell: applying causal discovery to single-cell analyses
Correlation between objects does not answer many scientific questions because of the lack of causal but the excess of spurious information and is prone to happen by coincidence. Causal discovery infers causal relationships from data upon conditional independence test between objects without prior assumptions (e.g., variables have linear relationships and data follow the Gaussian distribution). Causal interactions within and between cells provide valuable information for investigating gene regulation, identifying diagnostic and therapeutic targets, and designing experimental and clinical studies. The rapid increase of single-cell data permits inferring causal interactions in many cell types. However, because no algorithms have been designed for handling abundant variables and few algorithms have been evaluated using real data, how to apply causal discovery to single-cell data remains a challenge. We report a pipeline and web server (http://www.gaemons.net/causalcell/causalDiscovery/) for accurately and conveniently performing causal discovery. The pipeline has been developed upon the benchmarking of 18 algorithms and the analyses of multiple datasets. Our applications indicate that only complicated algorithms can generate satisfactorily reliable results. Critical issues are discussed, and tips for best practices are provided.