Filter features based on proportion

[JenKelly (Jennifer Kelly)]

Dear qiime developers,

I am interested in filtering features by proportion/fraction (i.e. 0.05% of total sample), however I can't find any way to do this.

I found this forum post stating that this functionality would be added in the future (Filter feature table based on percent prevalent?), could you let me know where I can find it.

Thank you,
Jen

[ben (benjamin w.)]

It looks like you will have to look @ your samples and determine how many sequences reach a 0.05% cut off and filter that way? This was sort of addressed on this thread:

Filter feature table based on percent prevalent?

Hello, I am interested in conducting low-prevalence filtering (e.g., 0.01%) on my feature table before doing diversity metrics or downstream analyses. The reason is that I am not interested in the rare biosphere, and want to focus on dominant taxa between two groups of samples. I cannot figure out how to do this using the feature-table filter-features function. Any help would be much appreciated. Thanks!

[Nicholas_Bokulich]

@JenKelly

I found this forum post stating that this functionality would be added in the future (Filter feature table based on percent prevalent?), could you let me know where I can find it.

Hi @JenKelly, It sounds like you want to filter by % on a per-sample basis (as opposed to % of all sequences). That functionality is still outstanding, as shown in this feature request. We will update you here when that feature is available.

[JenKelly (Jennifer Kelly)]

Hi @Nicholas_Bokulich and @ben

Thank you for your replies. Yes I am interested in filtering by % on a per-sample basis. Would a work around be possible as follows; convert table to relative frequencies, identify features below a certain relative frequency (of each sample), and then remove those from the original count table?

Firstly is it possible to filter features from a relative frequency table? And if so, how could I grab the feature names of those low abundance features?

Thanks for your help!
Jen

[Nicholas_Bokulich]

@JenKelly

Would a work around be possible as follows; convert table to relative frequencies, identify features below a certain relative frequency (of each sample), and then remove those from the original count table?

This would still not be per-sample filtering. It would only filter features that are less than a certain proportion in ALL samples, which you can do already without converting (figure out total # of reads and filter features that are observed less than some % of total reads).

@JenKelly

Firstly is it possible to filter features from a relative frequency table?

No.

If you want to filter on a per-sample basis, my guess is that you want to do this for plotting purposes. Is that correct? If you can describe the downstream goal I may be able to help guide you on an alternative route.