This repository contains the simulated data, the results of all the methods considered in the comparison, and the results of HATCHet (HATCHet repository) on the published whole-genome multi-sample tumor sequencing datasets, all these are described in the HATCHet paper at:

Simone Zaccaria and Ben Raphael, 2018

1. Simulated data
   • Tumors
   • Patients and samples
   • Results
     • Fixed
     • Free
2. Cancer data
   • Prostate cancer
   • Pancreas cancer
3. Analysis
   • Mutated copies, predicted VAF, CCF, and explaining of mutations

All the simulated data are included in the folder simulation. The simulated data comprises 256 mixed samples with 2-3 tumor clones for 64 patients (3-5 samples per patient), half with a whole-genome duplication (WGD). These simulated data have been generated using MASCoTE which has also been described in the HATCHet paper and is available here:

MASCoTE

Due to space limitations, we are unable to publish in this repository the sequencing reads from all samples. As such, for every sample we provide a BB file which encodes the read-depth ratio (RDR) and B-allele frequency (BAF) for every genomic bin of the reference genome in every sample. The BB files are produced by the pre-prosessing steps of HATCHet and summarize whole the basic input data needed by CNA-inference methods.

All these data are reported in the folder data. The patients are divided between those with a tumor without a WGD in noWGD folder and those with a tumor with a WGD in WGD folder. In both cases, a dataset corresponds to a collection of clones and is reported in the format dataset_nX_sYwhere X is the number of tumor clones (in addition there is a normal diploid clone) and Y is the random seed used by MASCoTE for reproducibility. The copy-number profiles of the tumor clones and the corresponding phylogenetic tree with CNAs and WGds are correspondingly reported in the two following files contained in the related subfolder tumor:

Each dataset includes two patients and for each patient a BB file describes the RDR and BAF of every genomic bin in all samples of the corresponding patient. The name of each BB file specifies the number of samples for the related patient, the number of clones, and the corresponding clone proportions. More specifically, the BB filename is given by a _-separated list where the first element preceeded by the letter k specifies the number of corresponding samples and each other element specifies the clone proportions of a sample, listed such that the first is the proportion of normal diploid clone and the clone proportion of any other tumor clone is given in corresponding order. The name of a sample is a _-separated list which starts with the noun bulk and each element specifies the clone proportion (without the dot) of every clone.

For example, k4_01090_02008_00506035_00504055.bb.gz is a BB file for a patient with 4 samples which incude 2 tumor clones (clone0 and clone1) and a normal diploid clone normal. In particular, the samples have the following clonal compositions

Another example, k7_040600_010090_020008_0103060_0205003_0100504_01030303.bb.gz is a BB file for a patient with 7 samples which incude 3 tumor clones (clone0, clone1, and clone2) and a normal diploid clone normal. In particular, the samples have the following clonal compositions

Each BB file corresponds to a patint and is a tab-separated file describing the RDR and BAF of every genomic bin in all samples in the following format:

Due to space limitations, each BB file has been compressed using gzip with level of compression 9. The file can be easily decompressed with the command gzip -d BBFILE.

HATCHet has been compared with 4 current state-of-the-art methods for CNA inference:

Each of these methods and HATCHet has been applied on the simulated samples. More specifically, Battenberg, TITAN, and THetA have been applied on each sample individually, cloneHD has been applied jointly on all samples from the same patient, and HATCHet has been applied both on each sample individually (single-sample HATCHet) and jointly on all samples from the same patient. We consider two different settings when executing the methods on simulated data.

First, every method has been applied on all 128 samples of the 32 patients without a WGD by providing the true value of the main parameters, including tumor ploidy, number of clones, and maximum copy number. In this case, the results obtained by every method are reported in the folder fixed and in the subfolder of the corresponding dataset. The results of Battenberg, TITAN, and THetA are specifically reported for every sample, the results of cloneHD are reported for every patient, and the results of HATCHet are specifically reported for every sample (when obtained by executing HATCHet on each sample inidividually) and specifically for every patient (when obtained by executing HATCHet jointly on all samples from the same patient).

Second, every method has been applied on all 256 samples of the 64 patients with and without a WGD, requiring that each method infers all the relevant parameters, including tumor ploidy and number of clones, and setting the maximum copy number to 8. THetA has been excluded from this analysis as it does not automatically infer the presence/absence of a WGD. In this case, the results obtained by every method are reported in the folder free and in the subfolder of the corresponding dataset, which are divided according to either the presence or absence of a WGD. The results of Battenberg and TITAN are specifically reported for every sample, the results of cloneHD are reported for every patient, and the results of HATCHet are specifically reported for every sample (when obtained by executing HATCHet on each sample inidividually) and specifically for every patient (when obtained by executing HATCHet jointly on all samples from the same patient).

For every method, all the most important and relevant output files are reported. The largest of these files have been compressed due to space limitations using the command gzip -9 and they can be easily decompressed by using the corresponding command gzip -d.

HATCHet has been applied on two whole-genome multi-sample tumor sequencing datasets; the first dataset comprises 10 prostate cancer patients analyzed in (Gundem et al., Nature, 2015) and the second dataset comprises 4 pancreas cancer patients described in (Makohon-Moore et al., Nature genetics, 2017).

The data for all prostate cancer patients are contained in the subfolder prostate. For each of the 10 prostate cancer patients (A10, A12, A17, A21, A22, A24, A29, A31, A32, and A34) the results inferred by HATCHet in a subfolder with the corresponding name. More specifically, the following files encode the results inferred by HATCHet for each prostate cancer patient:

The mutations inferred from all samples of every prostate cancer patient are reported in a subfolder mutations. The SNVs and small indels are reported in two comma-separated files indel_hc.csv and snv_hc.csv with the following fields

The data for all pancreas cancer patients are contained in the subfolder pancreas. For each of the 4 pancreas cancer patients (Pam01, Pam02, Pam03, and Pam04) the results inferred by HATCHet in a subfolder with the corresponding name. More specifically, the following files encode the results inferred by HATCHet for each pancreas cancer patient:

The mutations inferred from all samples of every pancreas cancer patient are reported in a subfolder mutations. The SNVs and small indels are reported in two comma-separated files indel_hc.csv and snv_hc.csv with the following fields

The section contains tools which have been applied for obtaining the analysis presented in the HATCHet's paper.

The tool for the analysis and clustering of SNVs computes and otuputs the following fields (specified in the header which starts with the symbol #):