I think this might need a little more description, with regard to the in_manifest id and instrument_run id. Are we assuming that the in_manifest id is the same simply because they came from the same study? If so, we probably shouldn't.

That first line of code is to confirm that all the runs have the same id, right? And then the next chunk confirms that the manifest_category is instrument_run. If I got that right, then I think my suggestion is to be more clear in the description above. Something like:

Look at the in_manifest id on these proteomic outputs to confirm that all runs are in the same manifest record, and pull that record. If that manifest record's manifest_category value is 'instrument_run', then it confirms that these are LC-MS/MS runs that were performed in succession on the same instrument. Proteomic outputs from different manifest records should not be aggregated.

Reply via ReviewNB

The code below does a little more than putting all the results in the same dataframe. In particular, I think more info about the last part would be good. You could just add more to the comment below, or add more to the explanation above. Either way, the type of protein becomes important below when doing FDR, so I think it's worth adding more info.

Determine the type of protein match for each peptide: contaminant, reverse (false positive match to the reversed amino acid sequence of a protein), or forward (match to the true, forward amino acid sequence of a protein).

Reply via ReviewNB

I am not wild about calling this 'noise'. Suggest the following change:

A challenge associated with aggregating mass spectrometry data is that there are always false identifications, which can be mitigated by imposing a spectral probability filter on the data being analyzed. The same spectral probability filter needs to be applied across datasets when they are being compared. The filter value itself is chosen by weighing the number of 'true' identifications retained with the proximity of the data to a chosen false discovery rate (FDR) (usually 0.05 or 0.01). NMDC's metaproteomic workflow provides 'true' and 'false' identifications for FDR estimation in the 'Unfiltered Metaproteomic Result' files.

Reply via ReviewNB

Collapse to unique peptides and normalize their relative abundance

Reply via ReviewNB

Would it be informative to show plots of the abundances both before and after normalization?

Reply via ReviewNB

I don't understand this step. Why would we sum abundance values?

Reply via ReviewNB

Edit:

Identify the razor protein, which is a method of limiting the assignment of degenerate peptides (i.e., peptides that map to more than one forward protein) to a most likely matched protein.

The rules are as follows:

• If a peptide is unique to a protein, then that is the razor
• If a peptide belongs to more than one protein, but one of those proteins has another unique peptide, then that protein is the razor
• If a peptide belongs to more than one protein and one of those proteins has the maximal number of peptides, then that protein is the razor
• If a peptide belongs to more than one protein and more than one of those proteins has the maximal number of peptides, then collapse the proteins and gene annotations into single strings
• If a peptide belongs to more than one protein and more than one of those proteins has a unique peptide, then the peptide is removed from analysis because its mapping is inconclusive

Reply via ReviewNB

Edit:

Perform sorted list protein mapping, which is a method of limiting the assignment of degenerate peptides to obtain a single protein identification for each peptide returned. This method does not use information from the original search protein fasta file, and thus is not the same as the popularly used 'first hit' strategy, although it employs a similar logic. It can be performed via the sortedproteins() function in agg_func.

It iss defined as:

• If a peptide is unique to a protein, then that is the sorted_list protein
• If a peptide belongs to more than one protein, but one of those proteins has a unique peptide, then that protein is the sorted_list protein
• If a peptide belongs to more than one protein and one of those proteins has the maximal number of peptides, then that is the sorted_list protein
• If a peptide belongs to more than one protein and more than one of those proteins has the maximal number of peptides, then the sorted_list is the first protein in a sorted list of all proteins across datasets
• If a peptide belongs to more than one protein and more than one of those proteins has a unique peptide, then the peptide is removed from analysis because its mapping is inconclusive

Reply via ReviewNB

Combine sortedmapping information with relative abundance values since sortedmapping returns a single protein for each peptide

Reply via ReviewNB

Starting here, I think we want to modify the rest of the notebook.

One thing I would want the notebook to do from this point, is to generate the input files needed for pMart. The table above is one, and the other would be the metadata table (rows with the sample identifiers and columns with the metadata factors like depth). Analysis of the combined data should be done in the software that we built for that purpose.

Gathering the annotation information for the proteins is useful, and so we should probably figure out where to put that information.

But searching across NMDC for other datasets with shared annotations or pathways is not something we want to do with this notebook.

Reply via ReviewNB