Restriction Fragment Handling¶
HiFive handles all data at the level of the restriction fragment via the
Fend classes. In the case of 5C data, this means the set of fragments that are specifically targeted by primers, whereas for HiC data, each restriction fragment is treated as two separate units, one for each half of the fragment. The reason for this is that each fragment end (fend) is associated with its own characteristics such as GC content and mappability. Further, because there is an inverse relationship between fragment length and overall interaction signal strength, successful ligation and sequencing is biased towards interaction sites (and therefore cross-linking sites) that occur closer to the end being ligated. Effectively this creates two different populations of interactions, those associated with each half of a restriction fragment.
All normalization and binning are accomplished from the fragment or fend level for 5C and HiC data, respectively. This means that once normalization is complete, it can be used for any level of resolution.
Loading 5C Fragments¶
Fragments associated with 5C data are loaded from a BED file that contains the chromosome, start and ending coordinates, name, and strand for each primer-targeted restriction fragment from the 5C experiment. Additionally, a FASTA file containing the primer sequences can also be provided to associate the GC content of each primer with its genome position data (used in HiFive’s Binning Algorithm). To reduce the storage space and processing time, only fragment data associated with primer targets are used in HiFive’s 5C Fragment objects, meaning that a different Fragment object is needed for each experimental design, but experiments sharing the same targeted fragments may share a Fragment object file.
Loading HiC Fends¶
Fends associated with HiC data can be loaded from either a BED file or a HiCPipe-compatible tabular fend file. If using a BED file, the file should contain either the chromosome, start, and stop position of each restriction fragment or the chromosome, start, and stop position of the restriction enzyme recognition sites for the target genome. HiFive will infer which type of BED file is given based on the coordinate intervals. If a HiCPipe-compatible fend file is given, it should contain a header line, fend, fragment, chromosome, coordinate and fragment length information for each fend in the target genome.
fend frag chr coord valid frag_len 1 1 1 3002506 1 3372 2 1 1 3005877 1 3372 3 2 1 3005878 1 389 4 2 1 3006266 1 389
HiCPipe-style fend files are 1-indexed, meaning that the first fragment and first fend both are labeled with a 1. This convention is used in HiFive only for these files to maintain compatibility with HiCPipe files.
The header line should contain the exact labels as seen above since HiFive uses them to determine which columns contain what information. In addition to the above characteristics, the tabular fend file may also contain the columns ‘frag_gc’ and ‘map_score’. These fend characteristic values are used in HiFive’s Binning Algorithm, although are not needed for either probability or express normalization.
Non-restriction Fragment-based Genome Partitions¶
As a consequence of chromatin interaction assay and analysis approach continuing to diversify, enforcing restriction enzyme-based genome partitions is too limiting. HiFive allows both uniform and arbitrary genome partitions for chromatin interaction analysis to support a variety of analysis needs.
There are currently three ways of creating ‘binned’ partitions, although they are all still stored in a
Fend object. These different approaches reflect differing experimental designs and care should be take to select the appropriate one.
Binned restriction fragment-based data
In order to support lower coverage datasets and facilitate analysis by users with fewer computational resources, HiFive allows binning of data as it is loaded from mapped read files. This requires knowledge of restriction sites as well as the binning interval. Thus a Fend object created for these kind of data should be created as described above for a standard RE-based partitioning but will also be passed a bin width via the ‘binned’ option. Datasets connected to the resulting Fend object will automatically be binned and all subsequent analysis will be on the binned data. This allows proper filtering of the reads but no need to have the overhead of keep fend pairs after this stage.
Uniform genome paritioning
If filtering based on RE-fragments is not needed, either because it has been previously performed or the experiment did not use a restriction enzyme, a Fend object with a uniform paritioning of the genome can be produced by passing a file containing only the chromosome names and lengths plus a bin width. Datasets associated with the resulting Fend object will not be able to perform filtering based on RE fragment assignment but can filter reads based on orientation as decribed for bin-based filtering. Bins produced in this approach always start from coordinate zero.
Arbitrary genome paritioning
HiFive also accepts user-defined paritions. The primary difference between these and restriction fragments are that unlike RE fragments, bins are no subdivided (i.e. fragments into fends) and no fragment-based filtering is performed on reads for datasets associated with these fend files. In order to tell HiFive that the parition file being passed is not restriction fragments, a value of zero must be passed to the ‘binned’ option.
Unlike RE-based paritions, within-bin interactions are included in binned partitions. This is all handled automatically by HiFive with one exception. If calling library functions directly, the ‘diagonal_included’ argument needs to be passed to plotting functions as they do not have access to the HiC object and therefore cannot determine whether the data are of the ‘binned’ type.