The Evidence Rank

The Evidence Rank indicates the quality of the annotations used at source to make the predictions. The Gene Ontology Dictionary comes with an evidence code for every annotation. If this evidence is good (for example, experimentally verified rather than electronically assigned), the reliability of the prediction made from such annotation is correspondingly higher.

GO evidence codes and their assigned ranks

	Evidence description	Code	Rank
1	Inferred by curator	IC	0
2	Traceable author statement	TAS	1
3	Inferred from direct assay	IDA	1
4	Inferred from mutant phenotype	IMP	2
5	Inferred from genetic interaction	IGI	2
6	Inferred from physical interaction	IPI	2
7	Inferred from sequence or structural similarity	ISS	3
8	Inferred from expression pattern	IEP	3
9	Inferred from Reviewed Computational Analysis	RCA	3
10	Non-traceable author statement	NAS	4
11	Inferred from electronic annotation	IEA	5
12	No data	ND	6
13	No record	NR	6

Please note that the evidence codes are arranged in descending order of their reliability. The order is somewhat arbitrary and merely reflects a belief that reliability of functional inferences observed from such evidences may be interpreted with the stated level of confidence.

The Evidence Rank is defined by the formula:

Evidence Rank (ER) is basically an cumulative average of the individual evidences (IE) of the annotations.

ER = IE_1, if only one GO annotation is used; ER = (IE_1+IE_2)/2 if two GO annotations are used; ER = (((IE_1+IE_2)/2) + IE_3)/2 if three GO annotations are used, and so on.

Please note that if the Evidence Rank is low, then it is a better value than high.
For example, an Evidence Rank with a value 5 would mean that the prediction was derived from all electronic annotations. Therefore even if the Bayesian weight of the prediction is high, the confidence in the result has to be interpreted with adequate caution.