Manual vs Blast2go annotation

upendra_35 · 01-30-2013, 01:35 PM

I have recently finished my transcriptome assembly and currently in the process of annotating ~50k transcripts. For annotating, i thought of using two approaches....

Blasting those transcripts against closely related well annotated sp (in this Arabidopsis) and pull out the GO terms for the best hits from the blast and assign those GOs to my transcripts.

Blasting those transcripts against Plant Refseq database and then use b2go pipeline to annotate them using default parameters in mapping and GO annotation stepS.

Now i am wondering how do i compare which of the annotation is best. These are the following i can think of:

a) How many transcripts that were annotated in the finally annotation.

b) How many transcripts that have GOs that are associated with "BP" term

c) Evidence Code distribution for hits and sequences of the GO terms for those transcripts.

What do you think of these criteria and what else i should be thinking of before deciding which annotation i should be selecting in the end.

mike.t · 01-30-2013, 08:48 PM

Hi -

You could calculate those statistics but unless you know what 'good values' are for the statistics you won't really be able to evaluate the performance of the annotation method. For example, you may ask, "How many transcripts were annotated?" and you may think that the method that annotates the most transcripts is the best, but in fact there could be a lot of false positive annotations. Incorrect annotations are as bad as no annotations when it comes time to interpret your results.

What you need are a set of similar transcripts that already have high quality GO annotations that you can use to evaluate different annotation methods. You can compare the results of the annotations that are output by the various methods and ask how similar they are to the original ones. There are a lot of different statistics you can calculate: precision, recall, sensitivity, specificity, etc. You could start by simply calculating the number of true positive and false positives and true negative and false negatives.

That said, if you have a closely related species, its pretty common to simply transfer the annotations from the best blast hit to your transcript. I'd be careful to use strict parameters in the blast search.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
blast2go command line!	PSW	Bioinformatics	1	12-03-2012 11:09 AM
Run Blast2Go offline	Slientinverse85	Bioinformatics	19	10-11-2012 06:43 AM
Help to use Blast2go	islandemiaj	Bioinformatics	3	07-23-2012 09:09 AM
Using Blast2go on a Mac	abrown1	Bioinformatics	0	05-25-2012 12:32 PM
Need ideas about the advantages of Blast2Go	byou678	Bioinformatics	5	08-27-2011 10:47 AM

01-30-2013, 01:35 PM	#1
upendra_35 Member Location: USA Join Date: Apr 2010 Posts: 96	Manual vs Blast2go annotation I have recently finished my transcriptome assembly and currently in the process of annotating ~50k transcripts. For annotating, i thought of using two approaches.... Blasting those transcripts against closely related well annotated sp (in this Arabidopsis) and pull out the GO terms for the best hits from the blast and assign those GOs to my transcripts. Blasting those transcripts against Plant Refseq database and then use b2go pipeline to annotate them using default parameters in mapping and GO annotation stepS. Now i am wondering how do i compare which of the annotation is best. These are the following i can think of: a) How many transcripts that were annotated in the finally annotation. b) How many transcripts that have GOs that are associated with "BP" term c) Evidence Code distribution for hits and sequences of the GO terms for those transcripts. What do you think of these criteria and what else i should be thinking of before deciding which annotation i should be selecting in the end.

01-30-2013, 08:48 PM	#2
mike.t Member Location: Spain Join Date: Mar 2010 Posts: 22	Hi - You could calculate those statistics but unless you know what 'good values' are for the statistics you won't really be able to evaluate the performance of the annotation method. For example, you may ask, "How many transcripts were annotated?" and you may think that the method that annotates the most transcripts is the best, but in fact there could be a lot of false positive annotations. Incorrect annotations are as bad as no annotations when it comes time to interpret your results. What you need are a set of similar transcripts that already have high quality GO annotations that you can use to evaluate different annotation methods. You can compare the results of the annotations that are output by the various methods and ask how similar they are to the original ones. There are a lot of different statistics you can calculate: precision, recall, sensitivity, specificity, etc. You could start by simply calculating the number of true positive and false positives and true negative and false negatives. That said, if you have a closely related species, its pretty common to simply transfer the annotations from the best blast hit to your transcript. I'd be careful to use strict parameters in the blast search.