Humberto Ortiz-Zuazaga humberto.ortiz@upr.edu

Finding differentially expressed genes in RNASeq from non-model organisms is a complex task. Usually a transcriptome is assembled from the reads, then the reads are aligned to the assembled transcriptome, then the counts are tested for differential expression.

Much of the reads are not differentially expressed, and assembly of non-model organisms is not trivial.

Can we pull regions of the assembly graph that are differentially expressed before assembling? Will it simplify the assembly?

This project seeks to answer these questions.

diffhash.Rmd R markdown code, uses the polyester package to generate simulated reads for a few genes, as demonstrated in the polyester vignette. Then calls the julia code below to find the occurrence of each kmer. With the count data, uses edgeR to find differentially expressed kmers and write them out to diffkmers.txt.

diffhash.jl Julia code to count the occurrence of each kmer in a collection of sequence files. Persists the counts in a JLD file. Uses BioSequences to read FASTA files and iterate over kmers.

showhash.jl Julia code to print the stored kmer counts.

Before removing kmers with fewer than 1 occurrence on average.

$ wc -l hashcounts.tsv
  964512 hashcounts.tsv

After filtering kmers

$ wc -l hashcounts.tsv
   36745 hashcounts.tsv

The result of running the .Rmd file is shown in diffhash.md.

We can include a read if any of it's kmers are differentially expressed or if all of it's kmers are differentially expressed:

any:

$ du -sh simulated_reads any
106M	simulated_reads
 48M	any

all:

$ du -sh simulated_reads filtered
106M	simulated_reads
 15M	filtered

Using "any" results in more false positives at this stage. All 20 of the simulated genes have at least one read with a differentially expressed kmer.

Real differentially expressed genes:

$ grep "^>" ../*.fa | cut -d\| -f 2 | head -n 4
424037187
424037186
209977002
52546690

Filtered set of genes (out of 20):

$ grep "read" *.fa | cut -d\| -f 2 | sort -u
209977002
213512645
29029549
29029551
424037186
424037187
52546690

That's 4 true positives and 3 false positives (out of 20 genes).

Use bcalm to build unitigs.

$ ls -1 filtered/sample_* > list_reads
$ bcalm -in list_reads -kmer-size 21 -abundance-min 2

produces a bunch of list_reads* files.

$ ~/src/spacegraphcats/bcalm/scripts/convertToGFA.py list_reads.unitigs.fa test.gfa 21

Bandage can read GFA and blast fragments against chr22_small.fa.

The 4 differentially expressed genes are the three structures on the left of the figure below, and the two small fragments on the right are small pieces of false positive genes.

1. Frazee AC, Jaffe AE, Kirchner R, Leek JT (2018). polyester: Simulate RNA-seq reads. R package version 1.18.0.

2. Robinson MD, McCarthy DJ and Smyth GK (2010). edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139-140

3. McCarthy DJ, Chen Y and Smyth GK (2012). Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation. Nucleic Acids Research 40, 4288-4297

4. Chikhi, Rayan, Antoine Limasset, and Paul Medvedev. "Compacting de Bruijn graphs from sequencing data quickly and in low memory." Bioinformatics 32.12 (2016): i201-i208.

5. Wick R.R., Schultz M.B., Zobel J. & Holt K.E. (2015). Bandage: interactive visualisation of de novo genome assemblies. Bioinformatics, 31(20), 3350-3352.