ATAqC

Sample Information

Summary

Note that all these read counts are determined using 'samtools view' - as such,
these are all reads found in the file, whether one end of a pair or a single
end read. In other words, if your file is paired end, then you should divide
these counts by two. Each step follows the previous step; for example, the
duplicate reads were removed after reads were removed for low mapping quality.

This bar chart also shows the filtering process and where the reads were lost
over the process. Note that each step is sequential - as such, there may
have been more mitochondrial reads which were already filtered because of
high duplication or low mapping quality. Note that all these read counts are
determined using 'samtools view' - as such, these are all reads found in
the file, whether one end of a pair or a single end read. In other words,
if your file is paired end, then you should divide these counts by two.

Filtering statistics

Mapping quality refers to the quality of the read being aligned to that
particular location in the genome. A standard quality score is > 30.
Duplications are often due to PCR duplication rather than two unique reads
mapping to the same location. High duplication is an indication of poor
libraries. Mitochondrial reads are often high in chromatin accessibility
assays because the mitochondrial genome is very open. A high mitochondrial
fraction is an indication of poor libraries. Based on prior experience, a
final read fraction above 0.70 is a good library.
  

Library complexity statistics

ENCODE library complexity metrics

The non-redundant fraction (NRF) is the fraction of non-redundant mapped reads
in a dataset; it is the ratio between the number of positions in the genome
that uniquely mapped reads map to and the total number of uniquely mappable
reads. The NRF should be > 0.8. The PBC1 is the ratio of genomic locations
with EXACTLY one read pair over the genomic locations with AT LEAST one read
pair. PBC1 is the primary measure, and the PBC1 should be close to 1.
Provisionally 0-0.5 is severe bottlenecking, 0.5-0.8 is moderate bottlenecking,
0.8-0.9 is mild bottlenecking, and 0.9-1.0 is no bottlenecking. The PBC2 is
the ratio of genomic locations with EXACTLY one read pair over the genomic
locations with EXACTLY two read pairs. The PBC2 should be significantly
greater than 1.

Picard EstimateLibraryComplexity

Yield prediction

Preseq performs a yield prediction by subsampling the reads, calculating the
number of distinct reads, and then extrapolating out to see where the
expected number of distinct reads no longer increases. The confidence interval
gives a gauge as to the validity of the yield predictions.

Fragment length statistics

Open chromatin assays show distinct fragment length enrichments, as the cut
sites are only in open chromatin and not in nucleosomes. As such, peaks
representing different n-nucleosomal (ex mono-nucleosomal, di-nucleosomal)
fragment lengths will arise. Good libraries will show these peaks in a
fragment length distribution and will show specific peak ratios.

Peak statistics

Naive overlap peak file statistics

IDR peak file statistics

For a good ATAC-seq experiment in human, you expect to get 100k-200k peaks
for a specific cell type.

Annotation-based quality metrics

Annotated genomic region enrichments

Signal to noise can be assessed by considering whether reads are falling into
known open regions (such as DHS regions) or not. A high fraction of reads
should fall into the universal (across cell type) DHS set. A small fraction
should fall into the blacklist regions. A high set (though not all) should
fall into the promoter regions. A high set (though not all) should fall into
the enhancer regions. The promoter regions should not take up all reads, as
it is known that there is a bias for promoters in open chromatin assays.

Comparison to Roadmap DNase

This bar chart shows the correlation between the Roadmap DNase samples to
your sample, when the signal in the universal DNase peak region sets are
compared. The closer the sample is in signal distribution in the regions
to your sample, the higher the correlation.