A dynamic retention time based support vector regressor (RT-SVR) model in
combination with q value assessment is described to improve the sensitivity
and accuracy of protein identifications.
It is specifically used for Mascot search results.
Input
To run RT-SVR, three input files obtained from Mascot database search are needed. How to obtain these files?
Mascot target output for model construction (FDR=0.01):
(Example)
Mascot target output for application at higher p value (e.g., p=0.10):
(Example)
Mascot decoy output for application at higher p value (e.g., p=0.10):
(Example)
q-value:
The default q value threshold is 0.01. Users can specify a q value threshold.
A recommended range for q value threshold is [0,0.10].
Output
In the output file, the predicted RT(shown as "Predicted"), RT error (shown as "Error") and q-value (shown as "q") will be appended as the last three columns.
Here is an Example output file by running RT-SVR using the above sample input files at q = 0.01.
Run RT-SVR
Download RT-SVR from here. Please read me before running RT-SVR.
RT-SVR_2.0 (with a combination of Mascot Identity Threshold (MIT))
With RT-SVR only to reevaluate Mascot peptide predictions, we found that around 5~10% of "good" predictions (peptide score >= identity threshold) are missed. To rescue these peptide identifications, we combine Mascot Identity Threshold (MIT) method with RT-SVR, by which a peptide identification will be marked as "valid" when its q value <= q threshold (e.g. 0.01) or its peptide score >= identity threshold. By this means, all "valid" peptide identifications will be obtained.
The requirements for installing and running RT-SVR 2.0 are the same as those for last version.
Download RT-SVR_2.0 from here. Please read me before running RT-SVR_2.0.
Last modified: May 25, 2011