uBiome Data Analysis Using MG-Rast

I've been meaning to write this for some time. This service is free to anyone who would like to peek a bit deeper into a uBiome report.  I wrote a similar post concerning American Gut reports a while back. This does not take an awful lot of skill to do, but you should try to be methodical and label your reports as you go, otherwise it will get messy a few months down the road when you add more.
 


Step-by-step Tutorial to load a uBiome Raw Data report into MG-Rast

1. First, get a free MG-Rast account.

2. From the uBiome dashboard, click on the "Download Raw Data" block, and save this zip file to your desktop or somewhere handy.

3. Open the zip file, you will see 8 files with a ".fastq.gz" extension. Save all 8 files it contains to your desktop or somewhere.  It helps to put an id in front of each file, ie. "Tim_ssr_19712_R2_L003_fastq.gz". 

[NOTE:  These 8 files are your sample, ran 8 times through uBiome's 4-channel sequencing machine.  They run the sample twice, forward and reverse, to look for all unique signatures that might be missed if running once in a one-channel machine.  So "R2_L003" is the reverse run, in lane 3.]

4. Log into MG-Rast, click the green "up" arrow in top right corner. 

5. In the upload area, click the "upload file" icon.

6. Select your 8 files, then "Start upload". When finished, "Next".

[NOTE: This can take minutes to hours, I believe you need to keep your browser here until they have all completed. Refresh the program periodically.  It will tell you when the upload is complete.]

7. Click "I do not want to supply metadata"

8. Click on "2, select project", easiest is to create a new project for each sample.  Just type in a project name such as "Joe's Guts", and select.

9. Click on "3, select sequence files."  Here you should see all 8 files you uploaded previously. If they are not there, you need to wait.  This can take minutes or hours.  Usually minutes for these type files (today it took 4 hours!).  You might need to go back through the metadata and project steps. But, you cannot proceed until you see all 8 files in your "inbox" in Section 3.

10. Highlight all files, click Select.

11. In Section 4, leave all the options as they are, hit Select.

12. In Section 5, click the top selection, "highest priority" and "Submit Job"

13. You'll get a confirmation pop-up and be linked to a status page.

14. As each file is processed, you'll get an email saying it is ready.  Takes hours or days, usually hours.

15. When you get the email, follow the links and you'll be able to see what was it that particular run. If you do not get an email in a day or two, log into MG-Rast and click the "Globe" icon in upper right corner and look to see if they are there.

16. In MG-Rast, you will eventually see all 8 files under the project name.  Click on the file, and it will pull up a page containing all sorts of information about your sample.

17. I find it helpful to open all 8 files at once, in order, and make some notes. You can spend hours analyzing your report from here, including the "diversity score" and also pie charts with taxonomy levels.  There are some other links at the top you can follow to further analyze the results.

Observations on MG-Rast vs uBiome Dashboard

There will be differences in each of the 8 runs, and none will precisely match the uBiome Dashboard.  Some runs will have very few "QC Errors" and others will have maybe 25-50% errors.  I always try to look at the runs that have the fewest QC errors.

This 8-run analysis shows the difficulty in getting out good reports.  uBiome seems to use their own formulas, different from MG-rast.  I think I trust the reports from MG-rast more than what uBiome gives because MG-rast compares your raw data with data banks from 6 different repositories.

uBiome takes a dab of your sample and applies it to an Illumina sequencing computer, generates a database of all the sequences seen, then also runs them in reverse to see if there are sequences that will match to ensure they get them all.

In a perfect world, all 4 forward runs would match exactly, and all 4 reverse runs would match exactly.  The two runs could be collated, and you'd see exactly what sequences are in your sample.

When you enter the data in MG-Rast, they are comparing the sequences in the fastq files with databases from 6 public repositories (RDP, Greengenes, SSU, LSU, ITS, and M5RNA). 

Here is a table I created showing my top 4 bacteria from a sample taken 2 July 2015. If you look closely, you can see all 8 runs are fairly balanced within a range.

(From Sample on 7/2/15...diet fortified with 4TBS of potato starch, genus level analysis)

Run/lane	Bacteroides      %	Fecalibacerium       %	Bifidobacterium       %	Blautia    %	Diversity  Score	Failed QC %
R1/1	27	18	19	4	19	27
2	28	20	11	5	20	28
3	23	13	14	4	19	23
4	29	8	25	3	21	29
R2/1	32	11	20	7	23	32
2	35	15	13	5	25	35
3	31	12	18	5	26	31
4	34	13	15	7	24	35

Here is the data on these bacteria provided by uBiome in my dashboard:

Run/lane	Bacteroides       %	Fecalibacterium       %	Bifidobacterium          %	Blautia    %	Diversity	Failed QC
uBiome	18	.74	16	.49	n/a	9.44%

As you can see, the information provided by uBiome was off by a huge factor in the Fecalibacterium and Blautia columns!  An email for an explanation has not been answered. This concerns me.

I've done 5 uBiome tests, and all of them show variances when compared to MG-Rast, but this report was the most disconnected. I tend to believe the reports provided by MG-Rast over uBiome. Fecalibacterium is normally present in quite large amounts in all humans.  I've looked at close to a hundred gut reports from uBiome, American Gut, and some others.  In none is Fecalibacterium less than 1%, especially in a high-fiber diet.

Possibly someone from uBiome can look into this and give us an explanation.

Until then...happy trails...
Hattip to Barney for helping me work through this tutorial!
Tim

Addendum:

(From Sample 7/2/15, phylum level analysis)

	Firmi	Bact	Actino	Eukary	Verruc	Proteo	Unc
R1-1	39	35	20	.9	1	.6	3
2	42	41	11	.8	1	1	1
3	31	49	15	.7	1	-	2
4	27	41	26	.7	1	-	1.4
R2-1	35	37	20	1.1	.6	.3	3
2	42	37	14	1.4	.6	.6	2.9
3	37	37	19	1.2	.4	.6	2
4	43	29	19	1.4	.7	.2	4.2
uBiome	19	52	23	1.7	.7	.3	?

 
Addendum 2:

Here is the analysis of a report dated 5/17/2015, all-potato diet, phylum level:

Run/lane	Firmi	Bacteroid	Actino	Proteo	Failed QC
R1/1	72	23	.9	-	4
2	76	21	.7	-	4
3	79	18	.9	-	4
4	77	20	1.1	-	4
R2/1	63	31	1	.6	12
2	70	26	1.1	-	14
3	61	33	1.1	-	14
4	60	34	1.1	-	15
uBiome	59	32	1.6	.25	6

And genus level:

Run/lane	Rumino	Bacteroides	Blautia	Roseburia	Eubacter
R1/1	12	20	5	.4	11
2	11	17	6	.4	9
3	14	15	6	.5	12
4	12	16	3	.4	10
R2/1	5	28	7	-	3
2	7	22	12	8	5
3	6	26	11	5	11
4	6	26	13	7	6
uBiome	12	11	7	3	.68

Just eyeballing these last two charts, you can spot statistical differences that require some explanation. For instance, the last column, "Eubacter" is Eubacterium, from uBiome's info page:

Eubacterium microbes are among the first to colonize us at birth, and even in our adulthood we carry them throughout our digestive, genital and urinary tracts. In our gut, these microbes are among the most frequent of the standard repertoire of characters, and it is thought that they help us digest resistant starches such as those found in beans, legumes, and unprocessed whole grains.

This bacteria is thought to be a normal commensal, beneficial bacteria. Part of Clostridium Cluster XIVa.  Read more here. If a doctor, or internet guru, was analyzing the uBiome report, they would think that this important species was missing, or very low, and recommend a high fiber diet.  As I was on a high fiber diet, and expected this bacteria to be much higher, I was surprised to see uBiome's report showing .68% Eubacterium.  However, MG-Rast shows it at a very normal 3-12% range, average about 8%.  Exactly what I would expect.