I'm trying to blast different nitrogen cycle related gene against various genomes got many hits with different E-value (starting from 0 to 9). What should be the threshold E-value. In literature, researchers used different thresholds for E-value, e.g. 1e-5, 1e-10, 1e-100 etc. I understand bit score, identity and other stats are also important. However, NCBI Blast can use only E-value as a constraint to research queries through the subject.
What should be a decent E-value?
E-value tells you the expected number of hits with a given score or higher.
At low E-values (between 0 - 1e-3), you typically use the numbers only to rank hits, e.g. to select the best one among many. Because otherwise, they are all significant (in the sense that they are not random noise). Hits with E-value > 1 are usually ignored, unless you know more specifically what you're looking for and that it may have very low similarity. In that case you must filter the hits that are satisfactory according to some other method (other than sequence similarity) or just take your chances and accept many false positives. Hits with intermediate values (between 1e-3 and 1) are borderline significant, how to treat those may be different from case to case.
In principle E-value lower than 0.05 can be considered as a statistically significant hit. However, in practice one consider even more stringent E-value cut-offs. A hit may have very low E-value but still can be a false positive. So you have to use your biology knowledge to judge its similarity/homology to your query sequence.
E-value tells you the expected number of hits with a given score or higher.
At low E-values (between 0 - 1e-3), you typically use the numbers only to rank hits, e.g. to select the best one among many. Because otherwise, they are all significant (in the sense that they are not random noise). Hits with E-value > 1 are usually ignored, unless you know more specifically what you're looking for and that it may have very low similarity. In that case you must filter the hits that are satisfactory according to some other method (other than sequence similarity) or just take your chances and accept many false positives. Hits with intermediate values (between 1e-3 and 1) are borderline significant, how to treat those may be different from case to case.
E-values are not a good measure for thresholding searching. They are dependent on the database, so if you change the database you get different e-values even if the matched sequences are the same.
BLAST is also a bad tool for looking at gene matches as it truncates sequences to get a better score. Always look at BLAST hits relative to the whole query sequence.
I agree with Frederico, a profile based method to be more sensitive and reliable.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics