Provide LC and CE retention time prediction values in output about metamorpheus

[trishorts]

Elude is hard to find. here it the github link to the code https://github.com/percolator/percolator/tree/master/src/elude_tool

[trishorts]

Best paper on the subject of LC prediction seems to be: https://www.ncbi.nlm.nih.gov/pubmed/28303880
which uses a support vector machine. An excellent lecture on SVR from an MIT professor (which greatly helps in understanding the paper) can be found here: https://www.youtube.com/watch?v=_PwhiWxHK8o

[trishorts]

one category
to learn on should be probability scores introduced for low res-bottom up

[trishorts]

I think that we should also consider adding and evaluating retention time predictions. This is standard for unmodifieds but I think would work well for modifieds. Delta retention times, scored by numbers of standard devations of all delta rts would be an excellent way to have an orthogonal measure of successful identification that would work well in percolator

[trishorts]

See (http://onlinelibrary.wiley.com/doi/10.1002/mas.21488/full) fore recent review of retention time prediction.

[trishorts]

ELUDE can predict retention times for modified peptides https://www.ncbi.nlm.nih.gov/pubmed/20735070

[stefanks]

Let's explore prediction time for mass shifts as well.

[stefanks]

Try to consider the number of modifications as a factor in the future implementation of percolator. Once that is done, the local FDRs should be indistiguishable, FDR should not be a function of anything.

[trishorts]

openMS also has some retention time prediction tools in C++ that can be found here https://github.com/OpenMS/OpenMS/tree/develop/src/topp

[trishorts]

LibSVM https://www.csie.ntu.edu.tw/~cjlin/libsvm/ has a C++ library with various pre-computed kernals. That might be something that we can use.

[zrolfs]

Another good (shorter) video for understanding SVR:https://www.youtube.com/watch?v=N1vOgolbjSc
Good packages for SVR are Accord and libSVM (libSVMsharp wrapper for c#)

[zrolfs]

I have Accord working.

[zrolfs]

Is this something we still want to have in the output? We can do it really easily with the existing libraries in mzlib, but it's not worth including if we don't think it'll be useful (or if it takes a long time for big searches).

[trishorts]

in a sense the calculation is already performed as it is part of PEP, but the value computed is not added to the output. Not ready for prime time b/c there is no consideration for CE and the computation for ptm modifiied peptides is crude. I assume the computation will evolve and when it gets somewhat more perfected, i'd still like it added.

Also, i think that rachel is adding it to metadraw, which will be nice.