Stpeter negative SIn value?

[sudarshan kumar]

and 

in StPeter what does negative SIn scor emean? I see all of them being negative 

[mhoo...@systemsbiology.org]

Here's a detailed mathematical explanation of SIn: https://doi.org/10.1038/nbt.1592

Briefly, the normalization process produces fractional representation of quantities, i.e. 0.01, 0.052, etc. These are then log-transformed, which results in values that typically range from -5 to -30, but are always negative because they are log-transforms of values less than 1.

These negative values can be adjusted by applying a scalar to the values (such as +30) so that all values are greater than 0. Alternatively, there is an option in StPeter (-s) where you can specify a total protein amount loaded into the instrument and StPeter will scale all the quantities to that amount. Note that this isn't an absolute value for each protein, because StPeter is standardizing the results to the sum of all observed proteins (not the ones you didn't identify), but it still puts all values in an easy to conceptualize, and positive, scale.

These standardized numbers are comparable across samples (to obtain ratios) so long as the samples were acquired with the same instrument method and chromatography conditions.

Cheers,

Mike

[sudarshan kumar]

OK, I can understand. so if the SIn values (displayed) are log transformed values. Can we take antilog of these values to come back to the actual fractional values.

Since, I want to compare two samples where i want to represent the difference in the quantity of each protein as fold change. Am I needed to do it separately in excel or other platform where i can calculate the ration using fractional (normalized SIn values) for each of the proteins.

Thank you 

[sudarshan kumar]

Hi Mike, thank you I did the way you said. Analyzed all samples separately.

I have a question

I kept on doing as

from mzml convert ---  comet search -- peptide prophet---- iprophet ---- protein prophet---- stpeter.

I didn't check the data quality neither filtered my data for anything during the process.

Can I do filtering in the end, based on protein probability as decided by the sensitivity and error table? I did filtering at the end choosing error rate of 0.05. 

I had 6 files of similar samples. I see that based on stpeter score the top hits (around 30 proteins) are similar across  6 samples. But the total number of proteins identified in each sample goes down drastically in few of the samples although all the files had similar number of displayed spectra upto iprophet.

CAn you please explain th reason?

thanks 

sud

--
You received this message because you are subscribed to the Google Groups "spctools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spctools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/af5ca34a-32b9-476e-98cd-185c21b6c483n%40googlegroups.com.

[mhoo...@systemsbiology.org]

Hi Sudar, 

I'm catching up here...

It looks like you've set your analysis pipeline properly - it is the same way I would do it. One recommendation, though, is to do the quantitation at an appropriate FDR threshold (i.e. FDR=0.01) instead of on all proteins. This is because SIn is normalized on all quantified signals. So including all protein groups means you are normalizing your quantities on proteins that are also not really in your sample.

As for the reason that some of your samples have fewer proteins than others, there are several possibilities, but they are unlikely to be related to StPeter. It could be, for example, that some of the charge state models failed in PeptideProphet, or that several high-probability decoy sequences skewed the distributions. Or perhaps a few key nondegenerate peptides were not observed, causing ProteinProphet to group proteins in drastically different ways. Or there are simply fewer proteins in those samples (number of spectra need not correlate to number of proteins). There isn't any one possible explanation without actually exploring the data, as each dataset will have its own characteristics that can influence the analysis.

Cheers,

Mike

[sudarshan kumar]

Ok Thank you. that makes sense. I should filter the list first before Stpeter quantitation.