Proline

Invalid Protein Set that don't have at least x peptides identifying only that protein set. The specificity is considered at the DataSet level.

This filtering go through all Protein Sets from worth score to best score. For each, if the protein set is invalidated, associated peptides properties are updated before going to next protein set. Peptide property is the number of identified protein sets.

Once pre-filters (see above) have been applied, a validation algorithm can be run to control the FDR. See how FDR is calculated

At the moment, it is only possible to control the FDR by changing the Protein Set Score threshold. Three different protein set scoring functions are available.

Given an expected FDR, the system will try to estimate the best score threshold to reach this FDR. Two validation rules (R1 and R2) corresponding to two different groups of protein sets (see below the detailed procedure) are optimized by the algorithm. Each rule defines the optimum score threshold allowing to obtain the closest FDR to the expected one for the corresponding group of protein sets.

Here is the procedure used for FDR optimization:

• protein sets are segregated in two groups, the ones identified by a single validated peptide (G1) and the ones identified by multiple validated peptides (G2), with potentially multiple identified PSMs per peptide.

• for each of the validation rules, the FDR computation is performed by merging target and decoy protein sets and by sorting them by descending score. The score threshold is then modulated by using successively the score of each protein set of this sorted list. For each new threshold, a new FDR is computed by counting the number of target/decoy protein sets having a score above or equivalent to this value. The procedure stops when there are no more protein sets in the list or when a maximum FDR of 50% is reached. It is has to be noted that the two validation rules are optimized separately:
  • G2 FDR is first optimized leading to the R2 score threshold. The validation status of G2 protein sets is then fixed.
  • final FDR (G1+G2) is then optimized leading to the R1 score threshold. Only the G1 protein sets are here used for the score threshold modulation procedure. However the FDR is computed by taking into account the G2 validated target/decoy protein sets.

The separation of proteins sets in two groups allows to increase the power of discrimination between target and decoy hits. Indeed, the score threshold of the G1 group is often much higher than the G2 one. If we were using a single average threshold, this will reduce the number of G2 validated proteins, leading to a decrease in sensitivity for a same value of FDR. In the future, we will try to implement such a strategy in order to allow the user to make its own comparison.