Title | | » | Use retention time regression to better predict peptide peaks over multiple replicates |
Assigned To | | » | Brendan MacLean |
Type | | » | Todo |
Area | | » | Skyline |
Priority | | » | 3 |
Milestone | | » | 1.2 |
In version 1.1, file AcquiredTime was added to the properties stored in the data cache. This made it possible to order the summary replicate plots by acquired time, when that differs from the order they are added in the document, e.g. document might be cases and controls, with multiple replicates per subject, but acquired in randomized order to avoid the confounding effects of acquisition order on peak area.
When many replicates are ordered by AcquiredTime, retention time from replicate to replicate tends to shift only very slightly, though it may drift by minutes over an entire set of dozens of replicates. It should be possible to do much better at peak picking by using some form of regression on retention time by acquisition order.
In this technique priority should also be given to the best scoring peaks in the entire set. i.e. if the 4 least intense concentrations in a calibration curve consistently cause the wrong peak to be picked, while the 3 most intense concentrations all pick a better peak, then the 3 should override the 4 worse peaks. Additionally, if the user actually manually chooses peak, it should be given highest priority, and all other peaks for replicates of the same peptide should be automatically re-assessed based on the picked peak. This would allow more flexible support for a technique in use in other packages, where the user is asked to manually specify the correct peak for all peptides in a training set.