The RiboHMM package can be installed through pip from the Github repository:
pip install git+https://github.com/djf604/RiboHMM
This will install two executables:
ribohmm, which is the interface for learning the model and inferring translated sequencesribohmm-utils, which is used to pre-process the input data and compute mappability
Aligned Ribo-seq and RNA-seq data (as BAM files) must first be converted to a tabix-indexed tabular
count format. The ribohmm-utils executable has a sub-program called bam-to-counts which automates
much of this process. Given multiple BAM files, bam-to-counts will output a single tabix-indexed
counts file which has aggregated counts from all input. For example:
ribohmm-utils bam-to-counts --bams sample1.riboseq.bam sample2.riboseq.bam --bam-type riboseq \
--output-prefix example_output
The above will read in sample1.riboseq.bam and sample2.riboseq.bam and produce two files:
example_output.ribo.counts.bed.gz, the tabular counts fileexample_output.ribo.counts.bed.gz.tbi, the tabix index of the above file
The sample BAMs can also be input directly into the main ribohmm executable and will be converted
as above on-the-fly.
For more detailed documentation refer to the wiki (coming soon).
Since ribosome footprints are typically short (28-31 base pairs), footprints originating from many positions in the transcriptome are likely to not be uniquely mappable. Thus, with standard parameters for mapping ribosome footprint sequencing data, a large fraction of the transcriptome will have no footprints mapping to them due to mappability issues. While RiboHMM can be used without accounting for missing data due to mappability, we have observed the results to be substantially more accurate when mappability is properly handled.
Given a GTF file that contains the transcriptome, mappability information (i.e., whether each position in the transcriptome can produce a uniquely mappable ribosome footprint or not) can be obtained in 3 steps:
-
For each desired footprint length (default 28-31bp), build a FASTQ with all footprints that could originate from the given transcriptome. This is done with the
ribohmm-utilssub-programmappability-generate.ribohmm-utils mappability-generate --gtf-file transcriptome.gtf --fasta-reference genome.fa \ --footprint-lengths 28 29 30 31 --output-fastq-stub example_outputThe above will produces four files:
example_output_footprint28.fq.gzexample_output_footprint29.fq.gzexample_output_footprint30.fq.gzexample_output_footprint31.fq.gz
-
Align the created synthetic FASTQs, using the same mapping strategy used for the original ribosome footprint profiling data. The BAM alignments will be the input into the next step.
-
For each desired footprint length, build a tabix-indexed mappability file from the BAM produced in step 2. This marks whether a footprint originating from a given position uniquely mapped back to the same place. This is done with the
ribohmm-utilssub-programmappability-compute.ribohmm-utils mappability-compute --mappability-bam example_output_length28.bam \ --output-tabix example_output_mappability_28.bedThe above produces two files:
example_output_mappability_28.bed.gzexample_output_mappability_28.bed.gz.tbi
For more detailed documentation refer to the wiki (coming soon).
The main interface to RiboHMM is through the ribohmm executable. By default, both the parameter
learning and inference steps are run in sequence. Some flags can be given to change that behavior:
--learn-onlycauses the program to learn the model parameters, save them to a model parameters JSON, and exit--infer-onlycause the program to skip the model learning step, instead accepting a model parameters JSON and moving directly to inference
For more detailed documentation refer to the wiki (coming soon).
In general the necessary inputs to learn the model parameters are:
- Reference genome in FASTA format
- Transcriptome in GTF format
- Ribo-seq BAM(s) or tabix-indexed counts file (created with
ribohmm-utils bam-to-counts)
Optional but helpful inputs include:
- Corresponding RNA-seq BAM(s) or tabix-indexed counts file (created with
ribohmm-utils bam-to-counts) - Mappability tabix-indexed counts files (created with
ribohmm-utils mappability-generate,mappability-compute)
An example:
ribohmm --learn-only --reference ref/genome.fa --transcriptome ref/transcriptome.gtf \
--riboseq-counts data/example_output.ribo.counts.bed.gz \
--rnaseq-counts data/example_output.rna.counts.bed.gz \
--mappability-tabix-prefix data/mappability/example_output_mappability \
--output run001 --batch-size 10
This produces a directory called run001 which has a single file model_parameters.json, which
can be passed into future --infer-only runs of RiboHMM.
Unless ribohmm is run with the --learn-only flag, this inference step is automatically run
following the learning step. This step can also be run directly by giving the --infer-only flag and
providing a model parameters JSON.
If run directly, this step generally needs as input the same set as the learning step, with the addition
of the a model parameters JSON, which is often called model_parameters.json.
An example:
ribohmm --infer-only --reference ref/genome.fa --transcriptome ref/transcriptome.gtf \
--riboseq-counts data/example_output.ribo.counts.bed.gz \
--rnaseq-counts data/example_output.rna.counts.bed.gz \
--mappability-tabix-prefix data/mappability/example_output_mappability \
--model-parameters run001/model_parameters.json \
--output run001
This produces inside of the run001 directory a file called inferred_CDS.bed, which contains
the inferred translated sequences.
If errors are encountered, please open an issue on this repository with a detailed bug report.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent