Predictions returning older EC labels, but pathways expecting the new labels?

[gbonthond]

Dear picrust2-team,

I am working on a 16S sediment dataset, and I know that it the samples are dominated by aerobes. With the v2.6.2 version, the aerobic respiration pathway (PWY-3781) is, however, not recovered. A while ago I used v2.5.3 on a subset of the same dataset and the PWY-3781 was very abundant. I know that the predictions have become much better since the PICRUSt2-SC, so I was surprised that a pathway i am certain is there, is not predicted anymore.

I noticed that some EC numbers have been re-labelled. For example, for the PWY-3781, the following ECs are needed:

EC 7.1.1.2 NADH:ubiquinone reductase
EC 1.3.5.1 succinate-ubiquinone reductase
EC 7.1.1.8 cytochrome bc1 complex
EC 7.1.1.9 Cbb3-1 cytochrome c oxidase

It seems that recently some of these have been relabeled in the kegg database:
EC 7.1.1.2 used to be EC 1.6.5.3 (https://www.genome.jp/dbget-bin/www_bget?ec:1.6.5.3)
EC 1.3.5.1 has still the same label
EC 7.1.1.8 used to be EC 1.10.2.2 (https://www.genome.jp/dbget-bin/www_bget?ec:1.10.2.2)
EC 7.1.1.9 used to be EC 3.6.3.14 (https://www.genome.jp/dbget-bin/www_bget?ec:3.6.3.14)

My final Picrust2 output that lacks the PWY-3781 pathway, also lacks the new EC 7.xxx numbers. I checked an my output does have the older versions (EC 1.6.5.3, 1.10.2.2, 3.6.3.14).

So in summary, I have the suspicion that the predictions of EC numbers still uses older labels while the mapping to the pathways assumes the newer labels. Could this be correct, and may that explain the absence of the aerobic respiration pathway in my dataset?

Thank you in advance!

Cheers,

Guido

[R-Wright-1]

Hi Guido,

I think that your explanation for the missing EC numbers/pathways could definitely be correct! I annotated the GTDB genomes using Eggnog v2.1.2, which gave me the EC numbers that we use in PICRUSt2. The files that map from EC numbers to pathways were taken from the default HUMAnN files. While the new database certainly improved predictions overall, I think it's always important to understand how the predictions are made and that they will always be limited by databases, as well as the overlap between databases. I am sure that I will get to updating the PICRUSt2 database at some point, but this takes a long time computationally. A couple of options for you:

• Manually edit your PICRUSt2 files to reflect the changes you described above.
• Use the instructions for creating a custom trait table (which will require you to either build of acquire a HMM for these functions)

I will also note that there are several other annotation options available in PICRUSt v2.6.2 - BiGG reactions, GO, gene names and PFAM (these are not run by default but can be run with the --in_traits option for the full pipeline). I don't know whether the pathway you are interested in is represented there at all, but this could be another avenue to pursue. I'd imagine that something like the gene names, in particular, wouldn't be likely to change as much as EC numbers.

Let me know if that doesn't make sense or if I can help more.

Robyn

[gbonthond]

Dear Robyn,

Thanks for your fast response! That is certainly helpful. So it is a mismatch between EC numbers outputted by the predictions and the EC numbers the pathway mapping takes.

In the meantime, I had a look at the older picrust2 output (v2.5.3), and here the EC numbers that are outputted are also the older ones. This means, the difference between the v2.5 and v2.6 versions is the mapping part (pathway_pipeline.py).

I also saw it is possible to run the pathway_pipeline.py step with -db DATABASE oldIMG. I will test this, and if it works, it will for now at least allow me to compare pathways of my new dataset with the older smaller version.

Thank you as well for the suggested options with "--in_traits". I look forward to a future updated version of picrust2 where the ECs from the predictions and those expected in the mapping step are matching :)

Thanks and have a great Easter!

Guido