The first-exon a-posteriori probability, P(exon), quantifies the probability -- computed by the quadratic discriminant function Exon-QDF described in the paper -- of finding a true first exon at the predicted location. A value of P(exon) = 1 means that the first-exon prediction is 100% correct, whereas a value of P(Exon) = 0 means that the first exon prediction is 100% incorrect.

The splice-donor a-posteriori probability, P(donor), quantifies the probability -- computed by the quadratic discriminant function Donor-QDF described in the paper -- of finding a true splice-donor at the predicted location. A value of P(donor) = 1 means that the splice-donor prediction is 100% correct, whereas a value of P(donor) = 0 means that the splice-donor prediction is 100% incorrect.

The promoter a-posteriori probability, P(promoter), quantifies the probability -- computed by the quadratic discriminant function Promoter-QDF described in the paper -- of finding a true promoter at the predicted location. A value of P(promoter) = 1 means that the promoter prediction is 100% correct, whereas a value of P(promoter) = 0 means that the promoter prediction is 100% incorrect.

By default, FirstEF prints out the predictions of all first exons that satisfy the three constraints:

P(exon) > 0.5,

P(donor) > 0.4, and

P(promoter) > 0.4.

This choice of cutoff values (0.5, 0.4, 0.4) results in a sensitivity and specificity of approximately 80%.

Advanced users may wish to run FirstEF with different cutoff values, in order to obtain more sensitive or more specific first-exon predictions. In principle, the user can adjust all three cutoff values independently of each other, which might be of interest in special circumstances where, for example, first exons with strong splice-donor sites and weak promoters or with weak splice-donor sites and strong promoters are searched for.

As a rule of thumb we recommend to modify only one parameter, for example P(exon), and choose the other 2 cutoff values proportional to P(exon), for example:

P(donor) = 0.8 * P(exon) and

P(promoter) = 0.8 * P(exon).

For example, if P(exon) = 0.5, we recommend P(donor) = 0.4 and P(promoter) = 0.4, or if P(exon) = 0.33, we recommend P(donor) = 0.26 and P(promoter) = 0.26.

Please note that cutoff values below 0.2 are not accepted, because the resulting predictions are not reliable for such small a-posteriori probability values.