xiFDR is an application for filtering crosslinked PSMs to a list of identifications with associated confidence values (False Discovery Rate).
The FDR can be calculated on several levels:
|PSMs||Just straight on the input – (crosslinked) peptide spectrum matches|
|Peptide Pairs||Turns PSMs into unique Peptide Pairs and performs the FDR estimation on Peptide Pairs|
|Proteins||Turns Peptide Pairs into unique Proteins and performs the FDR estimation on Proteins|
|Residue Pairs||Turns Peptide Pairs into unique Residue Pairs and performs the FDR estimation on Residue Pairs|
|Protein Pairs||Turns Residue Pairs into unique Protein Pairs and performs the FDR estimation on Protein Pairs|
One can either define to perform FDR estimation for only one or multiple desired levels (“FDR settings” Tab -> “Complete FDR”).
When several FDR levels are defined they will be applied in the order:
- First, the PSM level FDR.
- Second, the Peptide Pair level FDR
- Third, the Protein level FDR
- here, the peptide pairs get filtered to only contain those containing proteins that passed the Protein FDR
- Fourth, the Residue Pair level FDR
- Fifth, the Protein Pair level FDR
Additionally one can enable “boosting” on a given level:
When “boost” is selected, FDRs on lower levels are varied to maximise the number of observations on the specified level (e.g., on Residue Pair level these are Protein, Peptide Pair, and PSM levels that are varied).
If FDRs are additionally defined for the lower levels, these are set as an upper limit. Lower levels are considered superordinate to upper levels in this setup. Example: If both, 5% Peptide-Pair and 2% Protein level FDRs are selected, the boosting will increase up to a maximum of 5% on Peptide-Pair level before turning to boost Protein level FDR, even if this leads to a lower Protein level FDR than the one defined as upper maximum (i.e. < 2%).
- Ensure java version 8 (64-bit) is installed on your computer (Note: when downloading with a 32-bit browser an explicit selection of 64-bit download is required).
- For large-scale analysis, a computer with 16 GB memory is recommended.
- Operating system: Windows (Windows Server 2008 and 2012 and Windows 8 and 10) and Linux (Debian 9, Ubuntu 14.04, Ubuntu 16.04) were tested successfully. Any system running java 8 64-bit should suffice.