DEIsoM is a software using hierarchical Bayesian model for Differentially Expressed Isoform detection from Multiple biological replicates for identifying DE (Differentially Expressed) isoforms from multiple biological replicates representing two conditions, e.g., multiple samples from healthy and diseased subjects.
DEIsoM first estimates isoform expression within each condition by (1) capturing common patterns from sample replicates while allowing individual differences, and (2) modeling the uncertainty introduced by ambiguous read mapping in each replicate. Specifically, we introduce a Dirichlet prior distribution to capture the common expression pattern of replicates from the same condition, and treat the isoform expression of individual replicates as samples from this distribution. Ambiguous read mapping is modeled as a multinomial distribution, and ambiguous reads are assigned to the most probable isoform in each replicate. Second, DEIsoM couples an efficient variational inference and a post-analysis method to improve the accuracy and speed of identification of DE isoforms over alternative methods.
- A C++ compiler e.g. g++ 4.4.7
- boost v1.53.0+
- For previous versions, please remove -lboost_system in makefile
- Remember to set BOOST_HOME:
export BOOST_HOME=[path to boost home folder]
In the folder third_party, we have prepared the some third_party libraries:
- liblbfgs
- Used for optimization
./configureandmake. See its README for more installation instructions.
- samtools
- Used for reading bam files
cd htslib-1.3.1,./configureandmake.cd ..,./configureandmake. See its README for more installation instructions.
After correctly installing and configure boost, liblbfgs and samtools, we can install DEIsoM now.
First return to the main directory of DEIsoM and then build DEIsom by type make in command line for unix-like os.
After successfully compilation of the code, one can check its installation by testing the demo.
- Change the current working directory to
demo ./demo.sh- Check the output folder and kl.txt file. The kl.txt file will have the output like
chr22 ENSG00000100065 1 1 0.0912106
We need to prepare the annotation and seuqence files to sample and mapping reads. For simluation experiements, we used the gtf and gff3 files in Ensembl GRCh37 Release 84 Databases as the annotation files and the fasta files in GRCh37 patch_release_13 (GRCh37.p13) as the sequence files. For real HCC data, we used the gtf file in UCSC hg19 as the annotation file and the fasta files in GRCh37 Genome Reference Consortium Human Reference 37 as the sequence files.
As DEIsoM only accepts gff3 format annotation file. We could use the follwing script to conver gtf format to gff3 format:
gffread ensGene.gtf -o ensGene.gff3
For fast computing, DEIsoM first build indexed reference. We build a GFF3 file for every gene seperately from a single gff3 file. The annotation we used in our experiment can be downloaded from here. We now only support gff3 format. To build the indexed reference, one can run:
deisomIndexGFF [path to gff3 file] [output indexed reference folder]
- First, we can use alignment software to map the RNAseq data to our referece. Currently, we only support pair-end reads. There is no special requirements for the read length. We have tested tophat and RUM. And here is an example of command to run tophat:
tophat -p 1 -o [output folder] [reference folder] [pair-end read 1] [pair-end read 2]
- Then, we need to sort the accepted_hits.bam (the output of tophat) and make index for them.
samtools sort [path to accepted_hits.bam] -o accepted_hits.sorted.bam
samtools index accepted_hits.sorted.bam
- In our experiments, we used all mapped the reads which includes both the junction reads and non-junction reads.
Suppose we have M replicates and we want to build a model for N genes: gene 1, ..., gene N. The corresponding built indexed references are: gff 1, ..., gff N. We can run the model in one script:
deisomRun --gene-ids [gene 1], [gene 2], ..., [gene N] \
--gffs [gff 1], [gff 2], ..., [gff N] \
--bams [bam 1],[bam 2],...,[bam M] \
--read-len [read length] \
--paired-end [mean length of insert length] [standard derivation of insert length] \
--out-iter [number of iterations for updating alpha] \
--in-iter [number of iterations for updating variational parameters] \
--output [output folder]
The default output is in binary format. If you want to have a human readable format, you can add this option:
--human-readable
For computing the FPKM for the transcripts, we need you to pass an addition option:
--read-count [read count file]
Read count file is a file of M integers seperated by spaces. Each integer represents the total number of effective reads for each replicate. This can be computed using samtools. If you dont't want to compute the FPKM, this option can be omitted. FPKM for isoform k of a gene for the subject m is computed as
FPKM_{mk} = E[\psi_k] * numReadsMappedBySub[m] / numReadsPossible[k] / totalNumReadsBySub[m] * 1.0e9
For each gene. The run program will output two files: [Gene ID] and [Gene ID].fpkm.
If you use --human-readable option, the [Gene ID] will look like:
# isFinished
[1 or 0]
# gene
[Gene ID]
# numIsos
[K (# of Isoforms)]
# isoforms
[Isoform ID 1] [Isoform ID 2] ... [Isoform ID K]
# numSubs
[M (# of Subjects)]
# totalNumReadsBySub
[# reads in Sub 1] [# reads in Sub 2] ... [# reads in Sub M]
# numReadsBySub
[# reads mapped in Sub 1] [# reads mapped in Sub 2] ... [# reads mapped in Sub M]
# alpha
[alpha 1] ... [alpha K]
# betas
[beta 1,1] ... [beta 1,K]
...
[beta M,1] ... [beta M,K]
# rs
...
# done
The [Gene ID].fpkm file will look like:
# time
[running time]
# isFinished
[1 or 0]
# gene
[Gene ID]
# numIsos
[K (# of Isoforms)]
# isoforms
[Isoform ID 1] [Isoform ID 2] ... [Isoform ID K]
# numSubs
[M (# of Subjects)]
# totalNumReadsBySub
[# reads in Sub 1] [# reads in Sub 2] ... [# reads in Sub M]
# numReadsBySub
[# reads mapped in Sub 1] [# reads mapped in Sub 2] ... [# reads mapped in Sub M]
# FPKM
[FPKM for Isoform 1 Subject 1] ... [FPKM for Isoform K Subject 1]
...
[FPKM for Isoform 1 Subject M] ... [FPKM for Isoform K Subject M]
# done
Submitting a large number of genes using the above script is not efficient. And we sometimes want to run these jobs in parallel on a large cluster. We have provided a helpder program to automatically deivide the jobs in small chunks:
deisomSplit --path [path to deisomRun] \
--trunk-size [trunk size] \
--gff-dir [indexed gff directory] \
--bams [bam 1],[bam 2],...,[bam M] \
--read-len [read length] \
--paired-end [mean length of insert length] [standard derivation of insert length] \
--out-iter [number of iterations for updating alpha] \
--in-iter [number of iterations for updating variational parameters] \
--output [output folder]
You can also pass other options of deisomRun to it.
To evaluate the results, we can either use the KL divergence between alphas or the KL divergence between betas. We can use these commends:
- For general evaluations (either paired or unpaired between groups)
deisomKL [indexed GFF] [output of group 1] [output of group 2]
- For paired evalutions only
deisomKL --beta [indexed GFF] [output of group 1] [output of group 2]
The output of deisomKL will have multpile lines that looks like:
chr22 ENSG00000100065 1 1 0.0912106
Each line is an output for one gene. Column 1 is the chromosome ID. Column 2 is the gene ID. Column 3 represents whether the run program finished for group 1. Column 4 represents whether the run program finished for group 2. Column 4 is the KL divergence score.
We have upload the scripts and code for creating sythetic data in our paper. We also include all scripts for all other models that we used in the experiments. Please check simulation for details.
MIT License
Copyright (c) 2016 Hao Peng (pengh@purude.edu)
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Hao Peng, Yifan Yang, Shandian Zhe, Jian Wang, Michael Gribskov, Yuan Qi, DEIsoM: A hierarchical Bayesian model for identifying differentially expressed isoforms using biological replicates, in submission.