Detailed filtering power of order-1 SCFG model.

For each family, let the lowest score resulted by Infernal (or order-1 SCFG model) of all the known members (of all species) in the Rfam database be (or ). Setting the threshold to be (or ) would give the full power of eliminating false positives, but without omitting any existing real members. We define (or ) as an optimal threshold (Opt) for Infernal (or order-1 model). Yet to include possible novel members whose scores are lower than (or ), we also select two thresholds: and (or, and ) to compare the robustness of the filtering power of Infernal and our models. This table shows the detailed filtering power of order-1 SCFG model for the top 20 families in which Infernal reports the most number of false positives originally. As we can see in the table, the new order-1 SCFG model is able to filter more than 50% of those false positives reported by Infernal in all three thresholds. For the other families which are not shown in the list, the order-1 model can also filter over 50% of the false positives.