Identifying virus infection using human mRNA expression data

Question

[silje.nordgard]

I am interested in detecting the type of virus infection I may have in my human samples using available gene expression data from the same samples. Like; You have this signature, suggesting that your sample may be infected with virus X.

Does anyone have an idea how I best could do this?

I am afraid that what I have been doing so far merely suggests that I may have an infection, but doesn’t say much on the type.

Thank you for your input!

Answer 1

[GouthamAtla]

Blast/align the reads to ncbi viral genomes and see if you get any significant hits.

[silje.nordgard]

I only have the human expression data....

[GouthamAtla]

do u mean u have only expression values or sequencing data ? When you say "type of virus infection I may have in my human samples", you will have some viral sequences in your data, which could be detected by aligning them to viral genomes.

[silje.nordgard]

The expression data i have is not the raw data, it has already been mapped to the human genome. I have only the expression values for human genes.

[pld]

Yes, but you sequenced the mRNA within that sample. Depending on a number of factors, present viral material (mRNA, genomes, and so on) might have been picked up during sequencing. So you might have both human and viral reads, to find them you'll want to BLAST your reads against a viral genome database, as Goutham said.

With that said, depending on the sequencing approach, sample and what virus, there is large variability in how much and how well you'll detect virus. Huge. You also have to consider the difference between a pathogen and something present in the host microbiome. I would explore using PCR in your human sample to confirm, there are other methods 'wet' methods you might want to explore.

Answer 2

[h.mon]

Download the complete RefSeq viral database (or just the virus you are interested in), and use BWA or your preferred mapper to map RNAseq reads against this database.

edit: according to your comment, you either have counts of reads mapping to genes, or only reads mapping to genes (it is unclear to me which). Anyway, if this is the case, you can not say anything about viral status.

[Pierre Lindenbaum]

I would add: map all reads on the human REF , then extract the unmapped reads and map them on viral refseq.

[h.mon]

Indeed. Even better, map on human REF and viral REF simultaneously. BBSplit is a standalone tool for this, and FastQ_Screen and MGA are wrappers around Bowtie to achieve this as well.

[silje.nordgard]

Thank you for input. Unfortunately, the expression data i have is not the raw data, it has already been mapped to the human genome.

[h.mon]

What kind of data do you actually have? Aligned reads to the human genome in SAM or BAM files? Fastq files? Or text files with read counts? It would help if describe in more detail what you have done so far and what you have at hand now.

[silje.nordgard]

I have the counts of human transcripts per gene., like this:

ID      Sample1   Sampel2   sample3
Gene1   1         2         5
Gene2   1         0         4
Gene3   3         54        6
Gene4   2         4         8

[GouthamAtla]

Its not possible to tell the organism name/type by looking at numbers. Even if you have nucleotide frequencies, there could have methods existed, but by just looking at the expression table, its not possible.

[GouthamAtla]

If its already mapped to human genome, pull out the unmapped reads and map them to viral database. If you don't have sequence data, its nearly impossible to differentiate human and viral data.