Cellular DNA barcoding has become a popular approach to study heterogeneity of cell populations and to identify lineages with differential response to cellular stimuli. However, there is a lack of reliable methods for statistical inference of differentially responding lineages. Here, we used mixtures of DNA-barcoded cell pools to generate a realistic benchmark read count dataset for modelling a range of outcomes of lineage-tracing experiments. By accounting for the statistical properties intrinsic to the DNA barcode read count data, we implemented an improved algorithm that provides a significantly higher accuracy at detecting differentially responding lineages, compared to current RNA-seq data analysis algorithms. Building on the reliable statistical methodology, we illustrate how multidimensional phenotypic profiling (or high-throughput lineage phenomics) enables one to deconvolute phenotypically distinct cell subpopulations within a cancer cell line. The mixture control dataset and our analysis results provide a systematic foundation for benchmarking and improving algorithms for lineage-tracing experiments.

Improved detection of differentially represented DNA barcodes for high-throughput lineage phenomics