First Four Validations: DONE

By now we validate 4 things about the predicted genes: the length (by clusterization and ranking - the rank of the prediction among all the hits), the reading frame, whether there is a duplication in the prediction and whether the prediction is in fact a merge between multiple genes.

We adjusted our previous approach for merge detection in order to have fewer false positives. I'll briefly explain how we do the validation now:
- we plot a 2D graph by using the start/end offsets of the matched regions in the prediction on the two axis
- we draw a line obtained by linear regression 

- a gene merge is present if the slope of the line is between 0.4 and 1.2
- these thresholds were chosen empirically and will be adjusted after analyzing a larger amount of data.

Here it is a simplified drawing explaining this:


The output can be provided in html format (if you use the proper command line arguments). Here you can find sume examples, grouped by validation test:
Length validation:
Gene merge validation:
Duplication check: 

If I convinced you, try the application yourself:
Update: please clone from the developmenet branch
git clone
ruby prediction.rb -t protein --skip_blast --outfmt html gene_merge_validation/duplications/duplications.xml


Post a Comment