Uniquorn vignette

Raik Otto

2022-04-26

Uniquorn R package

Package to identify cancer cell lines (CCL)s based on their weighted mutational fingerprint.

1 How to make it work: Quickstart

Test run

Here the NCI-60 exome sequenced HT29 Cancer Cell line, reference genome GRCh37/ HG19

## Limiting out to top 3 hits per library.
## CELLMINER: HT29, matching variants: 665
## Correcting the background due to traces of random, scale-freeness amounts of matches, requiring at least 56 variants to match.
## Candidate(s): HT29786_0
## Storing information in table: /tmp/RtmpVMenx1/Rinst10f0f65a9feed9/Uniquorn/extdata/HT29.ident.tsv
##        CCL Matches   Library All_variants     P_values P_value_sig
## 1     HT29     665 CELLMINER          665 0.000000e+00        TRUE
## 2    786_0      56 CELLMINER          524 6.232112e-29        TRUE
## 4   SF_539      12 CELLMINER          340 7.097365e-03       FALSE
## 15 OVCAR_4       6 CELLMINER          312 3.432142e-01       FALSE
## 31    ACHN       3 CELLMINER          196 5.760371e-01       FALSE
## 17 COLO205       5 CELLMINER          343 6.013387e-01       FALSE
##    Above_Penalty Identification_sig
## 1           TRUE               TRUE
## 2           TRUE               TRUE
## 4           TRUE              FALSE
## 15          TRUE              FALSE
## 31          TRUE              FALSE
## 17          TRUE              FALSE

This will show a table with potential identification candidate, how many mutations overall and weighted of the training set have been found and if any training samples have surpased the identification threshold.

Explanation test data

The HT29 cancer cell line vcf file which contains the somatic mutations of the HT-29 cancer cell line was taken from Watson (not reachable anymore as of December 2016). The Watson repository contains the same cancer cell line samples as Uniquorn’s default training dataset, the original CellMiner panel, available @ cellminer. However, the cancer cell line vcf sample was differentely filtered and different algorithms have been used to predict mutations and variations. Therefore, it was used as example set to show the difficulties associated with an identification of cancer cell line sample: Some mutations are found only in the query version (here HT-29 from Watson) and some only in the training dataset (here HT-29 from CellMiner original panel). Moreover, some mutations of the query cancer cell line map to different cancer cell lines in the CellMiner training database what can lead to an incorrect identification of a cancer cell line.

Therefore, a robust yet sensitive cancer cell line identification algorithm is required.

You will find a file with the ending ’_uniquorn_identification.tab’ next to the input VCF file if you did not specify the output file path.

2 Add CCLE and CoSMIC CLP CL data

Please add the CCLE and CoSMIC CLP Cancer Cell Line (CL) data manually due to legal regulations! Else only the vanilla 60 CellMiner CLs will be available for identification. You can however manually add custom CLs.

‘CosmicCLP_MutantExport.tsv.gz’, unpack with e.g. gunzip on linux or 7zip on windows from http://cancer.sanger.ac.uk/cell_lines As of October the eights 2018, this is the first file listed, i.e. on top. If you cannot find the file, search for its name.

Important: utilizing the correct reference genome version is of utmost importance. In case of Cosmic CLP, ascertain that the websites’ reference genome version is correctly set.

‘CCLE_hybrid_capture1650_hg19_NoCommonSNPs_NoNeutralVariants_CDS_2012.05.07.maf’ from http://www.broadinstitute.org/ccle You can find this file in the ‘hybrid capture’ subsection

Registration for both websites is without charge and not complicated.

initiate_canonical_databases( ccle_file = 'path_to_ccle/CCLE_hybrid_capture1650_hg19_NoCommonSNPs_NoNeutralVariants_CDS_2012.05.07.maf', cosmic_file = 'path_to_cosmic/CosmicCLP_MutantExport.tsv.gz', ref_gen = "GRCH37")

One the initialization succeeds, about 2000 cancer cell line training sample for about 1200 different cancer cell lines are available in the Uniquorn’s database.

Note: Currently (January 2016), only the CoSMIC CLP data is available for the reference Genome version GRCh38. It is neccesary, that the reference genome for the training samples is specified if the version is not GRCh37

initiate_canonical_databases( ccle_file = 'path_to_ccle/CCLE_hybrid_capture1650_hg19_NoCommonSNPs_NoNeutralVariants_CDS_2012.05.07.maf', cosmic_file = 'path_to_cosmic/CosmicCLP_MutantExport.tsv.gz', ref_gen = "GRCH38" )

3 Add training CCL samples & utility functions

If you want to identify CL samples not contained in the ‘canonical’ CCL set, you can add your own custom CCL samples. These samples will be treated just as the ‘canonical’ training-datasets from e.g. CCLE. Note however, that it is strongly recommended to add at least 10 sample because overfitting might occur if too little custom training-samples are available.

add_custom_vcf_to_database( "path_to_file/my_own_CL_samples.vcf" )

Likewisely, if you want to remove the sample:

remove_custom_vcf_from_database( "Name_of_my_CL_custom_sample" )

If you want to see which CLs are contained:

show_contained_cls( ref_gen = "GRCH37" )

If you want to know which mutations are overall contained in the training set for a particular database:

show_contained_mutations( ref_gen = "GRCH37" )

Same if you want to know which genomic loci are associated with a particular CL:

show_contained_mutations_for_cl("SF_268_CELLMINER")

BED files and Broad Institute IGV visualization

Note as well, that there are BED files for the IGV Browser added as well, so that one can see the training, query and missed mutations in the genome. This feature can be switched of by setting the option output_bed_file in the identify_vcf_file function FALSE.

Contact: