This repository contains R functions implementing liquidCNA, a method for tracking emergent subclone dynamics in CNA data from longitudinal (liquid) biopsies. LiquidCNA leverages copy number data of many genomic segments tracked in multiple (cell-free DNA) biopsies over time accessible via low-cost shallow whole genome sequencing. The algorithm analysis the ensemble of segments and samples to derive an estimate of (i) the fraction of tumour-originating DNA (termed purity) of each sample; (ii) the fraction of tumour DNA originating from an emerging and potentially resistant subclone (termed subclonal-ratio) in each sampling time-point.
For further details on the method, please see our manuscript (Lakatos et al., 2021) in iScience.
All code has been written in R (version 4.0.3, compatible with alternative versions), with an example provided in Jupyter notebook format.
The following R packages are required for estimation or visualisation steps: pracma, ggplot2, ggpubr, reshape2, mixtools, fitdistrplus, dplyr, QDNAseq, gtools, gridExtra. Note that some of these packages are used only in pre-processing (e.g. QDNAseq) and might not be necessary for full function depending on the dataset.
The files in the repository are organised as follows:
-
mixture_estimaton_functions contains all functions of the estimation algorithm.
-
LiquidCNA_Example contains a detailed working example, using a synthetically generated dataset found in the Example sub-directory.
-
purity_estimate_synthetic and ratio_estimate_synthetic contain scripts automatically generating synthetic longitudinal CNA datasets (of 5 sampling time-points and 80 genomic segments) with varying levels of measurements noise. For each dataset, the scripts then run liquidCNA to estimate the purity and subclonal-ratio of samples, respectively; and results are recorded together with true values. ratio_estimate_synthetic_samplenumber is an extension of the ratio_estimate_synthetic script that implements random sampling of the number of longitudinal samples, then generates and analyses/estimates the thus sampled dataset. It also executes a bootstrapping solution for providing special care in the n=2 case (only 1 non-baseline time-point sample) which has higher uncertainty.
-
purity_estimate_insilico contains the script to automatically derive purity estimates for a sample set of 120 in silico cell line mixtures (obtained by sampling and mixing sequencing reads from two high-grade serous ovarian cancer cell lines and normal blood).
-
all_estimate_insilico automatically generates in silico datasets by randomly sampling samples from the 120 in silico cell line mixtures, including samples according to the minimum required read count (30 samples with >50M reads, 60 samples with >20M reads, etc.). Then the subclona-ratio estimate is computed for each sample, with purity values obtained from the estimates computed by purity_estimate_insilico. Results of samples purity (tumour fraction), relative and absolute subclonal-ratio are recorded together with true theoretical mixing proportions.
May 2021
Special cases and functions added for the n=2 (single non-baseline sample) case. This includes (1) updated estimation functions so dataframe operations do not throw errors and (2) a special bootstrapping function that allows random subsampling of subclonal samples prior to subclonal ratio estimation - to rectify that in n=2 unstable and subclonal segments are indistinguishable. Therefore prediction is now possible for a single non-baseline sample, but simulation on synthetic data show that the accuracy is lower and we would advise against.