I try to analyse similarity of short protein sequences (30-150 aa) to characterized proteins using NCBI Blast. Typical E-values of my queries range between 0.0001 and 0.97. Following an E-value definition at the NCBI website (see the link), would it be reasonable to consider all queries with E-value below 1 as potentially relevant to some biological function? It looks like less then one hit per query possible by chance alone is an acceptable situation, especially when a query with the top E-value of e.g. 0.3 has got more than one hit. If not, what threshold could be recommended? The goal is to shortlist biologically meaningful protein models. Thank you.
It really depends on what you want to show with your analysis. E-values take into consideration the size of your fragment and the database. For an E-value of 1, this is not really relevant, and means that you would expect this to have occurred by chance (An E of 1 means a match with this similarly value be expected at least 1 time using a database of this size and query of this size). If we look at the conversion of E-value to P-value:
"The BLAST programs report E-value rather than P-values because it is easier to understand the difference between, for example, E-value of 5 and 10 than P-values of 0.993 and 0.99995. However, when E < 0.01, P-values and E-value are nearly identical."
If we use the standard cut-off of P=0.05 to denote significance, you really shouldn't consider anything over about E=~0.05. Even then, this is only suggestive and you can't really draw too many conclusions, i.e. they may be related by evolution without conserved function, or not closely related but display convergent evolution. In my work, to denote that two proteins could be homologous, I generally use an e-value cutoff of e-5 or -10.
If you wanted to increase the likelihood of reaching lower e-values you can decrease the size of your database. If you are using a small protein sequence and blasting this against a large database like the "nr" non-redundant, it would be very difficult to reach significance.
If you are using this information to predict structural motifs, you may consider using the score instead, which relates to the overall quality of the alignment, not the probability of obtaining the alignment. A high score would indicate the sequences are highly similar.
Dear Michael, thank you very much for a very clear and comprehensive explanation. I analyze alternative ORFs. Your answer gave me a good idea as to how much I can rely upon the similarity alone. Good luck with your own studies!
set up a cut-off say 1e -3 or -5 or -10 and check your results if anything positive hit when you go from -10 to -3. If nothing change take -10 if more meaning full hits with e-3 then go for it. Again totally dependent on strategies and objective of the work.
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology