Estimating Gene Count Talk

Estimating Gene Count

How many reads are overlapping genomic features? - or - Can we confidently assign each read to a feature/transcript/gene? Not so simple.

We also have:

• Multi mapping reads
• Overlapping genes/transcripts

Two approaches:

• Focus on what’s known with certainty
• Probabilistic

Multi mapping reads

• Unsolved problem:

- this can account for 10-30% of reads

• Ignore them, but then again this decreases sensitivity
• Weighted assignment

Of course, longer reads would solve this problem.

One transcript, one set of reads

Two transcripts, another set of reads

[[File: t1t2.png

Third transcript, another set of reads

HTSeq-count

• Designed for RNA-Seq counting
• Simple to use (especially since v0.6.0)
• Work at gene level
• Remove multi-mapped reads
• Several modes to resolve remaining uncertainty

HTSeq-count

Probabilistic approach

Cufflink

cuffdiff

Probabilistic approach

Cufflinks: Reconstruct the transcripts from the data and annotation

Probabilistic approach

Cufflinks: Reconstruct the transcripts from the data and annotation

Cuffdiff: Assign each read/fragment to a transcript with a probability maximum likelihood.

Probabilistic approach

Cufflinks: Reconstruct the transcripts from the data and annotation Pros: - Better methodology - Integrated package (ease of use) Cons: Cuffdiff: - Do not support alternative experiment design - History of heterogeneous results/versions

• Assign each read/fragment to a transcript with a probability maximum likelihood.