I want to screen some proteins for a human protein using BLASTp. I want to find a unique protein from my query sequences of protein that are non homologous to human proteins. What should be the expected threshold for it (default is 10)? Should I go with 0.0001 (expected threshold)?
It depends on the confidence you require in the result; the E-value can be chosen to define this. With an E-value threshold of 0.01 when you do 100 queries you can expect on average one of them to contain a false result by chance. How many errors can you tolerate in your results?
If you are only interested in non-human results then you can improve your chances very slightly by removing all human sequences from the dataset first.
Thanks julian gough for your informative answer. when i did the blastp with three different E value 10 (default), 0.01,and 0.0001, i got hits (non human protein) 2, 25, 70. [data set contain 170 sequence).
Can you help me with which e-value should i go for my remaining 4100 protein sequences.
An E-value is per-query. So your errors are also per query. BLAST (or any other program) will work out the correct E-value for your query. It will adjust based on your data set size.
It sounds like the query you tested was 1 sequence versus a dataset of 170. It is not clear whether the 4100 are (a) sequences you want to search against your dataset of 170, or (b) whether you want to add them to your data set of 170.
If it is (b) then just add it to the dataset and rerun. If it is (a) then if you repeat teh search 4100 times then your errors will add up, and so you should adjust your E-value accordingly. I.e. to ensure a 1% chance of an error in your overall result, use an E-value threshold of 0.01/4100. BLAST probably has a way to search two sets of sequences against each other and *should* adjust the E-values for you (in which case you search 170 vs 4100 and use an E-value of e.g. 0.01). I can't tell you how, I no longer use BLAST because it is no longer the best method for anything. I recommend JackHMMER or the HHsuite from Soeding's lab, which includes CS-BLAST and others. BLAST is a historical method.
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Proteins