Protein inference

To allow protein group comparison, groups were created using the combined peptide evidence of all compared samples. Two different protein grouping methods were tested: MPA²⁸ and PAPPSO⁶¹, and analyses were made on protein groups and subgroups (Supplementary Note ³).

Assigning peptides to their correct protein can be a difficult task, notably due to the protein inference issue³, i.e., the same peptide can be found in different homologous proteins. This is particularly challenging in metaproteomics where the diversity and number of homologous proteins are much higher compared to single-species proteomics. To overcome this issue, most bioinformatic pipelines tend to automatically group homologous protein sequences into protein groups. However, each tool handles protein inference and protein groups in its own way, which prevents a straightforward output comparison at the protein group level. In order to allow robust comparison between approaches, the PSM output files of the four bioinformatic pipelines were combined. The peptides were then assigned to protein sequences in the FASTA file and the data was prepared for subsequent protein grouping. Two approaches of protein grouping were used and evaluated in this study: PAPPSO grouping⁶¹, which excludes proteins based on the rule of maximum parsimony, and grouping from MPA²⁸, which does not exclude proteins. All data processing was done using a custom Java program except for PAPPSO grouping for which data was exported and imported using the appropriate XML format.

For both methods, protein groups were created using the loose rule “share at least one peptide” (groups) and the strict rule “share a common set of peptides” (subgroups), resulting in a total of four protein grouping analyses: (1) PAPPSO groups, (2) MPA groups, (3) PAPPSO subgroups, and (4) MPA subgroups. Finally, the resulting protein groups and subgroups were exported for further analysis (Supplementary Note ³). These algorithms are also implemented in Pout2Prot⁹⁵ for independent use.