Thursday, November 08, 2007

Idiots look at Data, Part III: Slightly closer look

Having identified what data we're going to look at in Part II, we're going to start zooming in. This process may appear painfully slow, but I'm posting things as I put them together, and I'm not spending much as much time as you might think pre-analyzing things to fit conclusions you might imagine I've already drawn. To some degree I'm going to follow ideas suggested in the comments.

However, I do have a general intention of looking for the forest, not the trees, even though we appear to be looking closer and closer at individual branches.

Anyway, in this installment, I'm just going to put up narrower views of the graphs shown in Part II without a much of my own analysis. I've tried to crop them all to the relevent parts, and scaled to a 640 pixel wide image. In some cases, the tops of peaks are truncated. Also at the end are the Shackleton chromatograms from GDC-1101.

[MORE]


Click on any image for larger size.



Figure 1: UCLA GDC 1362/ WMA slides Page 40

No reported values available. 4 labeled peaks, plus about 8 not called out if you squint; 12 total. By far the cleanest.


Figure 2: Ex 92 Landis F3

Ex 92 - April testing of 993865 - LNDD 1398 (Landis from 3-Jul). There is no reported value for the 5aA, with no explanation given. It could be a decision to say it has matrix interference. 5bA reported as -1.04.

16 peaks labeled by LNDD, plus at least 20 not called out; 36 total.


Exhibit for for 993856 from 11-Jul not available, reported as 5aA -2.91 and 5bA at -4.09



Figure 3: Ex 88 Landis F3

Ex 88 - April testing of 993855 - LNDD 1015 (Landis from 13-Jul). 5aA reported as -4.62, 5bA as -4.09.

8 peaks labeled by LNDD, plus 25 or so not called out; 33 total. Second best Landis on IsoPrime2.



Figure 4: Ex 90 Landis F3

Ex 90 - April testing of 825425 - LNDD 1207 (Landis from 14-Jul), reported as -1.01, -0.70

16 peaks labeled, plus at least 20 not called out; 36 total.



Figure 5: Ex 86 Landis F3

Ex 86 - April testing of 825428 - LNDD 824 (Landis from 18-Jul). Reported as -5.06, -3.56

22 peaks labeled, plus at least 25 not called out; 47 total.


Figure 6: USADA 173, Landis Stage 17 A sample F3.

USADA 173, reported as 5aA -6.14, 5bA -2.15, Andro (other fraction) -3.99. Done on the IsoPrime1 machine with different software, so the display is different.

7 labeled peaks, plus at least 22 not called out; 29 total (note Isoprime1).

Figure 7: USADA 349, Landis Stage 17 B sample F3

USADA 349, the Stage 17 B sample, reported as 5aA -6.39, 5bA -2.15, and Andro -3.51. Also done on the Isoprime1 with different display.

7 labeled peaks, plus at least 20 not called out; 27 total (note IsoPrime1); Best Landis on Isoprime1.




Figure 8: Ex 87 Landis F3

Ex 87 - April testing of 825429 - LNDD 922 (Landis from 22-Jul). Reported as -5.06, -3.56

22 labeled peaks, plus at least 17 not called out; 39 total.



Figure 9: Ex 84 Landis F3

Ex 84 - April testing of 825424 - LNDD 633 (Landis from 23-Jul). Reported as -4.96, -1.45.

17 labeled peaks plus at least 15 not called out; 32 total. Best Landis on IsoPrime2.



Figure 10: Aquilera control F3

[UPDATE: earlier version of this post duplicated Fig 9 here]

Ex 93 - April testing of 825427 - LNDD 1492, the ocean blue, (Aguilera control), reported at 5aA -1.31, 5bA -0.95

11 labeled peaks plus at least 10 not called out. 22 total. One of the best.



Figure 11: Aquilera control F3

Ex 85 - April testing of 825426 - LNDD 730 (Aguilera control), reported as 5aA -0.77, 5bA -0.88

20 labeled peaks plus at least 15 not called out; 35 total.



Figure 12: Aquilera control F3.

Ex 89 - April testing of 825423- LNDD 1193 (Aguilera control), reported at 5aA -1.21, 5bA -0.79

16 labeled peaks plus at least 20 not called out; 36 total.


Figure 13: This is Figure 3 from the Shackleton paper in GDC-1101.

The top graph in Shackleton has 4 labeled peaks, and 13 not called out, 17 total, looking pretty clean.

The second has 5 labeled peaks, and 14 not called out, 19 total, looking pretty clean.

Summary

Test
Labeled
Non labeled
Total Peaks
quality
comment
UCLA
4
8
12
good

Ex 92 3-Jul
16
29
35

No 5aA reported. Why?
Ex 88 13-Jul
8
25
33
2nd best Landis on IP2

Ex 90 14-Jul
16
20
36


Ex 86
22
25
47


USADA 173 20-Jul
7
22
29
worst IP1

USADA 349
20-Jul
7
20
27
best IP1

Ex 87 22-Jul
22
17
39


Ex 84
23-Jul
17
15
32
best Landis on IP2

Ex 93 control
11
10
22
best on IP2

Ex 85 control
20
15
35


Ex 89 control
16
20
36


Shackleton top
4
13
17


Shackleton bottom
5
14
19






5 comments:

m said...

TBV,

I think I figured it out.

The UCLA slide is the etio and andro which elutes at about 1200+ seconds, etio followed by andro. This is the Landis F2. So you should compare the UCLA slide with the F2 graphs such as USADA 0343.

You will see that the andro and etio peaks while close together seem pretty well defined in the Landis F2 at USADA 0343. Yes there is a little more noise in areas away from the etio and andro, and one big peak following the andro. On the other hand blowup of the UCLA graph shows that the etio peak trails into the small following andro peak.

You should not be comparing the UCLA slide to the F3 graphs because it appears the 5B and 5A graphs are just more noisy, maybe because it's just harder to do the separation chemistry for those metabolites.

One proper comparison is the bottom graph in the Shackleton figure 3 above, which shows an intervening peak between the 5B and 5A and some possible shoulders. That's pretty messy by your criteria, yet Shackleton found it adequate for his carbon ratio calculations in his pioneering study. And I seem to recall Shackleton testified that the LNND chromatograms were fine.

DBrower said...

M,

when I look at details, I think the Shackleton chromatograms are still very much better, in a quantifiable way, that I'll be working through.

The observation on the UCLA slide is a point well taken, and I need to think about whether it affects my analysis. I think I am looking at a different kind of meta-analysis, and it may not matter, but I'm not sure.

TBV

m said...

TBV,

One other thing. I don't think total peaks is a good metric. What we are concerned about is the noise around the metabolite peaks of interest. Noise far away doesn't affect our measurements.

DBrower said...

The goodness of a metric depends on what you are using it for, and how reliably it tracks what you are concerned about. I can accept that what I'm doing is not what M is immediately concerned about, but that is not my goal.

TBV

Mike Solberg said...

For the record, TbV, there is a reason given why there is no value given for the 5aA in the July 3 test. On page 92 of exhibit 92, it says:

"NB: La valeur isotopique du 5aA n'a pu etre determinee en raison de sa faible concentration dans l'echantillon."

"Note bene: the isotopic value of 5aA was not determined for the reason of its weak concentration in the sample."

Not that it matters now. But I cannot see from exhibit 92 why it has any weaker concentration in the sample than in any other sample. It looks to me like they just couldn't decide if it was peak 14 or 15 in the July 3 IRMS 'gram, so they punted.