emase 0.10.5

EMASE: Expectation-Maximization algorithm for Allele Specific Expression

EMASE: Expectation-Maximization algorithm for Allele Specific Expression

Narayanan Raghupathy, Kwangbom Choi, Steve Munger, and Gary Churchill

• Free software: GNU General Public License v3 (GPLv3)

• Documentation: https://emase.readthedocs.org.

Note: The documentation for EMASE is still under work.

What is EMASE?

EMASE is a software program written in Python to quantify allele-specific expression and gene expression simultaneously from RNA-seq data. EMASE takes in the diploid transcriptome alignment BAM file and GTF file as inputs and estimates expression abundance for each isoforms and each alleles using Expectation Maxmization algorithm.

Why Use EMASE?

Current RNA-seq analysis pipeline employ two steps to quantify gene expression and allele-specific expression (ASE); gene expression is estimated from all read alignments, while ASE is assessed separately by using only reads that overlap known SNP locations.

Large-scale genome sequencing efforts have characterized millions of genetic variants across in human and model organisms. However development of tools that can effectively utilize this individual/strain-specific variation to inform quantitation of gene expression abundance have lagged behind.

EMASE, together with Seqnature (https://github.com/jaxcs/Seqnature), offers an integrated solution to utilize known genetic variations in quantifying expression abundances at allele and gene/isoform level.

In F1 hybrids from model organisms, EMASE allows us to utilize parental strain-specific genetic variation in RNA-seq analysis to quantify gene expression and allele-specific expression (ASE) simultaneously

In humans, EMASE allows us to utilize the individual’s genetic variation in doing personalized RNA-seq analysis and quantify gene expression and allele-specific expression (ASE) simultaneously

Briefly, EMASE: EM for allele-specific expression, uses individualized diploid genomes/transcriptomes adjusted for known genetic variations and quantifies allele-specific gene expression and total gene expressionsimultaneously. The EM algorithm employed in EMASE models multi-reads at the level of gene, isoform, and allele and apportions them probabilistically.

Earlier, we developed Seqnature to utilize known genetic variations, bot SNPs and Indels, to build indivdualized genomes and adjust annotations. One can use Seqnature to create individualized diploid transcritpme and align RNA-seq reads simultaneously to the diploid transcriptome and get alignment file in BAM format. This diploid BAM file can be used as input to EMASE.

Applications

Allele-specific gene expression in F1 Hybrids from model organisms

If we have F1 hybrids with parental genetic variants information, one can use Seqnature to build strain specific genomes and extract diploid transcriptome. RNA-seq alignment bam file obtained by aligbning RNA-seq reads to the diploid transcriptome is used as input for EMASE.

Personalized ASE analysis in Human

EMASE can be used to do personalized RNA-seq analysis in human. For this, we use the phased genetic variation (SNP and Indel) information to construct personalized diploid genome and align reads to the diploid transcriptome..

Allele-specific Binding using Chip-Seq in F1 Hybrids

Although we explained the use of EMASE to quantify Allele-Specific Expression from RNA-seq data, the tool can be used with other types of sequencing data. We have successfully used EMASE to quantify allele-specific binding from ChIP-seq data. While useing ChIP-seq, one needs to use diploid binding target sequences instead of diploid transcriptome for alignment target sequences.

Mining Diploid alignments and alignment probabilities

EMASE can be used to glean more information from the alignment, in addition to running EMASE and obtaining effective read counts for each allele and isoform. For example, we can use EMASE’s count-alignment program to obtain unique reads at allele-level for every gene, gene unique reads but allele-level multireads, and the total number of reads aligned to every gene. Having these alignment statistics at for every isoform and gene can be useful in interpreting expression estimates from EMASE.

References

• EMASE: Expectation-Maximization algorithm for Allele Specific Expression, Narayanan Raghupathy, Kwangbom Choi, Steve Munger, and Gary Churchill, Manuscript in preparation.

• [RNA-Seq Alignment to Individualized Genomes Improves Transcript Abundance Estimates in Multiparent Populations](http://www.genetics.org/content/198/1/59.short) Steven C. Munger, Narayanan Raghupathy,Kwangbom Choi, Allen K. Simons, Daniel M. Gatti, Douglas A. Hinerfeld, Karen L. Svenson, Mark P. Keller, Alan D. Attie, Matthew A. Hibbs, Joel H. Graber, Elissa J. Chesler and Gary A. Churchill. Genetics. 2014 Sep;198(1):59-73. doi: 10.1534/genetics.114.165886.

• [PRDM9 Drives Evolutionary Erosion of Hotspots in Mus musculus through Haplotype-Specific Initiation of Meiotic Recombination](http://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1004916) Christopher L. Baker, Shimpei Kajita, Michael Walker, Ruth L. Saxl, Narayanan Raghupathy, Kwangbom Choi, Petko M. Petkov, Kenneth Paigen PLOS Genetics: published 08 Jan 2015 | info:doi/10.1371/journal.pgen.1004916

History

0.10.5 (2016-01-20)

• Added ‘pull_alignments_from’ method in AlignmentPropertyMatrix class

• Added a script ‘pull-out-unique-reads’ that unsets emase pseudo-alignments that are not uniquely aligning

0.10.3 (2016-01-06)

• Fixed a bug in run-emase on handling inbred (reference or one haplotype) alignments

0.10.2 (2016-01-04)

• Added get-common-alignments: To compute intersection between each of paired ends

0.9.10 (2016-01-04)

• AlignmentMatrixFactory can handle unmapped reads

0.9.8 (2015-07-31)

• Fixed a bug in simulate-reads: No more duplicate read ID’s

0.9.7 (2015-07-28)

• Added create-hybrid: Build hybrid target directly using custom transcripts

• Added simulate-reads: Four nested models

0.9.6 (2015-06-02)

• AlignmentPropertyMatrix can represent Equivalence Classes

• Fixed a bug in length normalization

• Swapped Model ID’s between 1 and 2

  • Model 1: Gene->Allele->Isoform (*)

  • Model 2: Gene->Isoform->Allele (*)

  • Model 3: Gene->Isoform*Allele

  • Model 4: Gene*Isoform*Allele (RSEM model)

0.9.5 (2015-05-17)

• Fixed length normalization: Depth = Count / (Transcript_Length - Read_Length + 1)

0.9.4 (2015-02-23)

• Fixed a bug in ‘prepare-emase’

0.9.3 (2015-02-22)

• Fixed a bug in Model 2 of handling multireads

• ‘run-emase’ checks absolute sum of error (in TPM) for termination

0.9.2 (2015-02-17)

• Added three more models of handling multireads

  • Model 1: Gene->Isoform->Allele

  • Model 2: Gene->Allele->Isoform

  • Model 3: Gene->Isoform*Allele

  • Model 4: Gene*Isoform*Allele (RSEM model)

0.9.0 (2015-01-31)

• First release on PyPI

• Only implements RSEM model for handling Multiply-mapping reads (or multireads)