PaSD-qc ("Power Spectral Density-qc", pronounced "passed-qc") provides comprehensive evaluation of the properties and qualities of single cell whole-genome sequencing (scWGS) data.
Anyone who works with scWGS libraries knows that the technology is a fickle beast. Sometimes the whole-genome amplification works great. Sometimes it works terribly. Biases abound, and these biases can affect the accuracy of variant calls. A central challenge is thus to determine the quality of amplification prior to deep sequencing and to measure the biases prior to variant calling.
PaSD-qc provides a simple command line tool to rapidly evaluate the amplification properties and quality of scWGS sequencing using a custom, highly accurate power spectral density estimation technique. PaSD-qc provides:
You can read more about the powers of PaSD-qc in our paper at Nucleic Acids Research and by exploring the examples directory of this repository.
PaSDqc is currently in beta. A stable release will be available soon.
git clone https://github.com/parklab/PaSDqc.gitPaSDqc directory and run make
Preprocessing bam files: PaSD-qc requires indexed BAM files aligned to either the GrCh37 or hg19 reference genomes as input. To ensure consistency across samples, we recommend sequencing (or downsampling) all libraries to the same depth prior to running PaSD-qc. Depth recommendations:
Quick usage: PaSDqc QC -d <dir/of/bams> -o <my/output/dir> -c db/categorical_spectra_[XX].txt
db/categorical_spectra_[XX].txt is the gold-standard spectra corresponding to the depth of your samples (see FAQ for more)Long usage: PaSDqc QC [OPTIONS]
Options:
-h display help message and exit
-i list of bam files to analyze [bam1 [bam2 ...]]
-f file containing paths to bam files to analyze (instead of -i, -d)
-d directory to search for bam files to analyze (instead of -i, -f)
-o output directory (will be created if it does not exist)
-c categorical spectra file
NOTE: if using generic spectra, be sure to specify the correct depth (see FAQ).
-n number of threads to use
-b Build {grch37, hg19}
-q Mapping quality for read extraction
-r name of html reportEven longer useage PaSDqc -h
positional arguments:
extract Extract depth at uniquely mappable positions from bam file(s)
PSD Estimate power spectral densities from coverage files
report Generate HTML report of single-cell sample quality
QC Run all PaSDqc steps, starting with BAM files and ending with an html reportPaSD-qc is comprised of three steps: 1) read depth extraction; 2) power spectral density estimation; 3) report generation. These steps can be run independently by the user or all at together by using the QC method.
Why should I used PaSD-qc?
PaSD-qc offers unparalleled insight into the properties of single-cell whole-genome sequencing. While other QC approaches generally provide information only about read depth dispersion (often at a fixed-genomic scale), PaSD-qc provides comprehensive quality control. For example, PaSD-qc can infer the full distribution of amplicon sizes in a sample. This provides a principled method to determine what bandwidth to use when employing a single-cell specific caller which uses allele balance of nearby loci to perform accurate variant calling.
How do I interpret the results in the html report?
Where can I find more data on my samples not included in the HTML report (e.g. chromosomal amplicon properties)?
In the output directory, PaSD-qc creates a sub-directory data. With this directory you will find
How do I know if my sample is high or low quality?
The answer to this question depends on the application, so unfortunately no single answer exists. In general, however, the more closely a single-cell PSD resembles a bulk PSD, the higher the quality. That is, total variance should be small; sub-amplicon variance should approach that of bulk sequencing; and the lower the supra-amplicon variance, the better. Additionally, the smaller the variance of the amplicon size distribution, generally the higher the quality of the sample. Most importantly, the PSD of all samples should be highly similar. Any sample with a noticeably different PSD should be discarded as low quality.
Additionally, PaSD-qc can compare the behavior of new samples to previously calculated gold-standard spectra. PaSD-qc is distributed with generic gold standard-spectra for depths corresponding to 1x (db/categorical_spectra_1x.txt) and 0.1x (db/categorical_spectra_0.1x.txt). However, the shape of high-quality PSDs is highly dependent on amplification kit and protocol, so we recommend that you generate your own gold-standard spectra based on your own samples which you believe to be high quality. Instructions for how to do this can be found in example/08_example_gold_standard_spectra.
If you need help interpretting the results of your analysis, please feel free to email us (contact info at bottom of page).
Can I use PaSD-qc to evaluate whether sub-chromosomal regions are poorly amplified?
Yes. See examples/07_example_region_blacklist for instructions.
What should I supply as a bulk sample with the -u flag?
The -u flag is optional. By default, PaSD-qc will use a generic bulk PSD to normalize your single-cell samples. This will work fine for general purpose evaluation of samples. However, if you performed bulk sequencing at the same time as you sequenced the single cells, you may get more accurate estimates by using your own bulk sample. To do this, simply run PaSD-qc on the bulk sample and then supply the path to the resulting PSD file using the -u flag when profiling the single cells. E.g
PaSDqc QC -u /path/to/your/<bulk_name>.chroms.spec ...I deleted the html report. Do I have to rerun the whole pipeline?
No. The steps of PaSD-qc can be run separately. To regenerate the report, simply run ./PaSDqc.py report -d /path/to/analysis/dir -c categorical_spectra_XX.txt. Regenerating a report generally takes less than 1 minute.
To see all functions of PaSD-qc, run ./PaSD-qc.py -h.
Do you have some example data I can try out on?
Yes. The source code is distributed with an examples directory, which contains comprehensive instructions on basic and advanced usage of PaSD-qc. The tutorials are distributed as Jupyter notebooks, which can run once PaSDqc has been installed. To check that your installation is functional, you can re-compile the report in 01_example_report navigating to the cloned repository and running
PaSD-qc.py report -d examples/01_example_report -c categorical_spectra_0.1x.txtThis will recreate the html report in the directory examples/01_examples_report. If you want to start with BAMs, the samples used in our paper are available on SRA. See the Availability section of the paper for accession numbers.
Doesn't plotly upload its plots to the interwebs? Can anyone see my PaSD-qc plots??
Rest easy, none of your plots are uploaded to the internet. We use the plotly javascript API to embed the plots directly into the html report. This means that a report is entirely self-contained, so you can send it around to collaborators without fear of plots not loading. It also means that the reports can be fairly large (several MBs), as the data required to generate the plots must also be embedded in the html file.
Can I used PaSD-qc to evaluate my single-cell whole-exome sequencing?
In theory, yes, the approach should work on whole-exome sequencing. However, the method to perform the power spectral density estimation should probably be adapted to use information about the targeted capture regions. We are eager to do this. If you have high quality single-cell WES data and are interested in PaSDqc analysis, please contact us. We would be happy to work with you to ensure the results are high-quality.
I want to use PaSD-qc to analyze the quality of samples aligned to a different genome build. Is this possible
Yes, but you'll need to a bed file of uniquely mappable positions for the genome. Contact us for help with this.
Maxwell Sherman: maxwell_sherman {at} hms.harvard.edu
Peter Park: peter_park {at} hms.harvard.edu