Sunday, September 23, 2007

Discussion with BustinBilly on matching GCMS with GC-IRMS

In the Saturday Roundup, BustinBilly made a comment:

Don't debate the retention times just look at the damn plots. They are printed @ UsadaAndLandis-FinalAward20-09-07[1].pdf pages 41, 42. They are exactly the same. No doubt.

I used to believe Floyd until I saw the evidence.

To which I replied:
Excellent, not [meant "now"] we can have some useful discussion.

I'm sure you understand that only the F3 fractions are relevant; why the others were included is a mystery.

Your commend say the plots are "exactly the same"

In what ways do you think they are the same? Once you identify some of those we'll have something to talk about.

thanks!
TBV

And BB replied:
The F3 is the 5-alpha-androstandiol-5beta-pregnandiol difference this whole shebang is about. The A sample F3 is @ pages 35, 36. The GC/MS and GC/C/IRMS plots are identical. The two metabolites above and the 5-beta-androstandiol are the 3 peaks in the middle of the GC/C/IRMS plot.

with clarification

I meant the 3 big peaks in the middle.

Because this is important, I've given this discussion it's own post here, and expect to carry on in comments below. Warning: I will be merciless deleting off-topic comments.




I believe BB is referring to the the following chromatograms. First the GCMS of the blank, the Landis F3, then the two overlaid to show what Brenna testified, the retention times line up. This was never disputed.


Figure 1:
Blank F3 GCMS plot, labeled by compound; very clean chromatogram.
click for full size
USADA-345

Figure 1a:
Landis F3 GCMS

labeled by compound
click for bigger
USADA-348



Figure 1b:
Blank F3 and Landis F3 GCMS overlaid
Scaling isn't perfect, but the identified compounds line up within TD2003IDCR by time.

(Note the very messy and sloping background in the Landis compared to the blank, and the extra peaks it contains.)

And then the GC-IRMS plot:

Figure 2:
Landis F3 GC-IRMS chromatogram,
labeled by retention time and CIR.
click for full size
USADA 348

To make things clear, below are versions that put the things on top of each other, going along with BB's (and USADA/Majority) claim that the big peaks in the middle are absolutely the ones in question. That is, one is lined up, and the scale adjusted linearly so the other peak lines up as well. This anchors the highest middle peak, and adjusts the scale so the rightmost big peaks line up. These are the "peaks of interest" for the measurements.

Figure 3:
blank F3 GCMS and Landis F3 GC-IRMS
chromatograms overlaid.
click for full size


Figure 3a:
Landis F3 GCMS and GC-IRMS overlaid.

Asserted -28.79/5bA aligned, scale adjusted so 5bPreg matches.
click for bigger.

To make sure I've understood BB's claims, I'll ask now, did I get this right? Please answer in comments below and we'll carry on the dialogue. When needed, we'll add more figures to this post, and if BB wants to add some, we'll include whatever he provides.

Billy writes:

There you have it. The three peaks are the same.

I reply:

OK. Remember what we did, we adjusted the horizontal scale to make those three peaks line up, so we think linear (proportional) adjustment is correct.

Now, look at figure 4, a zoom of the left hand side of figure 3a, where the "anchor" internal standard is supposed to be located.

There doesn't appear to be very good alignment with those peaks is there?

With the linear scaling we used to get the peaks in the middle and right to align, we don't have alignment on the peaks on the left.

How should we try to correct it, by changing the scaling factor, or shifting the left peaks into better alignment?

Figure 4:
zooms in on Figure 3a.
click for bigger

Figure 4a:
Zoom in on figure 3.

Click for bigger.

If these zooms aren't clear, please suggest what I can try to make them better. Alternatively, try these manipulations yourself with Photoshop/Gimp and the previous figures. I do it by making one a background layer, and the other a transparent layer on top, which can be scaled and moved around.

Billy Answers:

There is nothing to correct. It is a qualitative issue.


tbv@trustbut.com said...

By what do you mean "qualitative?"

Should the internal standard align in the GCMS and GC-IRMS the way the analytes of interest line up or not?


BustinBilly said...

It is qualitative. For identification only. Which is different from a quantitative issue, e. g. the T/E quantification that was negated. The identification of the three metabolites is quite obvious. What more is there to say?


docnzym said...

But the internal standards are suppose to be lined up or why would you call them an internal standard (other than for extraction efficiency purposes) By just looking at these three peaks and nothing else then it would seem that you are forcing the answer


BustinBilly said...

What does it matter if the quantified peak-internal standard difference is changed by 0.3-0.4 minutes? The patterns are the same. The peaks were identified in the GC/MS plot.

The plots are from different machines with corresponding columns. A difference is expected.


BustinBilly said...

Look, there are two pertinent points I haven't seen anybody argue away.
1.The substances were identified correctly in the F3 GC/MS plots.
2.The F3 GC/C/IRMS plots for both the A&B samples are nearly identical to the GC/MS plots. The three metabolites of interest are almost exactly identical.

The panel of judges was right when they decided this issue.


TBV adds some figures:

Figure 5:
Lined up on the SI peaks on the left, scaled horizonally so the 5bP lines up.
This is interesting.
Click for bigger.
Figure 6:
Scaling from Figure 4a slid left so SI peak aligns.
5bP on right is way off.
Click for bigger



Figure 7:
Zoom of Figure 5 (not 6) showing 5bPdiol peak alignment
I believe the peaks align.



Figure 8:
Zoom of Figure 5 (not 6) showing 5bPdiol baseline match.
It's not as good as the peak which is why both are shown.



Figure 9:
Zoom of Figure 5 (not 6) showing baseline in the middle.


TBV saaid


[...] responding to BB's 5:45 comment.

1. The GCMS identification isn't in question; that's not in dispute here.

2. This is what we are discussing, and we'll see if it gets argued away or not.

In the comment that Billy started with, he said "They are exactly the same".

We've shown with overlays they are not exactly the same, and Billy has changed to using "nearly identical" for the whole chart and saying the metabolites of interest are "exactly identical".

He says the lack if exact match on the low end near the SI is not relevant. We'll leave this aside for now.

I do want to offer two more versions of overlays for comparison.

Figure 5 is a version of the overlay where the SI is aligned, and the axis linearly expanded so the 5bP aligns. In this, we see a significant left shift of the 5bA and 5bP peaks.

Figure 6 is one where the overlay scaling of Figure 4a is left shifted so the SI seems to align. In this case, it doesn't appear anything on the middle or right aligns, so that might be pointless to pursue.

Figure 7 is a zoom of the 5bP peak [in Fig 5] to show how it was aligned at the top. Figure 8 shows that the bottom alignment might be questioned, but the trailing peaks look OK.

Figure 9 is a zoom [in Fig 5] of what we think is the region of interest. In it, things are not quite as clear to me as they appear to be for BB. Here we see the peak identified as 5bA might align as well with a low GC-IRMS peak as with the one originally claimed, and the one labelled 5aA might align with the GC-IRMS peak that is ostensibly the 5bA.

Are the identifications of the GC-IRMS peaks still as obviously correct on visual examination as they seemed at first?


BustinBilly said...


Yes the identifications are as obviously correct as they seemed at first.

A good analogy would be different sized fonts. If you type the same sentence in the same font in Photoshop, the two sentences will overlay perfectly. Type the same sentence again in the same font-using a slightly larger point. A few letters will line up well on overlay, the rest will not. However it will still be the same sentence.

I hope that analogy helps. I don't see how quibbling over the overlay makes any difference.


tbv@trustbut.com said...


Please be patient while I try to figure out what is obvious to you, and not yet to me.

In Figure 9, we can see that innaccurate horizontal alignment can result in peaks being completely misidentified. The GCMS peak for 5bA is closer to a small GC-IRMS peak just short of 15:00 minutes that has not had it's CIR measured. The GCMS peak identified as the 5aA peak nearly sits on top of the GC-IRMS peak that is measured to be the 5bA peak.

How is that potential misidentification not a problem?

If that misidentification can't happen, what criteria are you using to decide which peak in that region aligns with which peak in the other chromatogram?

[ There's some discussion I'll capture in a bit, but for now, moving on ]



Figure 11:
Landis A sample F2 GCMS
USADA 165
click for bigger



Figure 12:
Landis A sample G2 IRMS
USADA 167
click for bigger



Figure 13:
Figures 11 and 12 overlaid, and scaled so anchor and rightmost labeled peak align.
click for bigger.


64 comments:

BustinBilly said...

The GC/MS plot you have posted is the blank not Floyd's sample. See the Misc Info reads blanc urinaire. You need the F3 plot that reads 178/07 995474. I can't tell what GC/C/IRMS plot that is because there is no identifying info.

Don't know if you will be able to overlay the plots because the scales are different. Looks like 2 minutes on GC/MS=3 minutes on GC/C/IRMS.

DBrower said...

OK, I will check up on the GCMS and correct when I get a chance.

That the scales are different is precisely the point of using relative retention times; given a known anchor and the RTs each of the analytes, in the GCMS and IRMS from calmixes one can factor out the scale differences.

I believe the essence of your claim is that the general shape of the peaks is the same so those are the right peaks, and I'm trying to provide a visual example of that claim. Let's not run forward till what I'm showing agrees with what you're saying is visually obvious.

TBV

DBrower said...

Also, while I go dig for the right GCMS, the locations of the peaks is supposed to be the same in the blank and the sample F3 GCMS, so I don't think this affects the substance of the argument. I will correct Figs 1 and 3 to remove all doubt.

TBV

DBrower said...

Corrections are in, with original figures relabeled to be correct, and new figures added with different numbering.

I have not paid much attention to the vertical scaling, as it isn't really relevant. The horizontal scale has been adjusted with a linear interpolation. The -28.79/5bA peaks were aligned, and then the horizontal scale adjusted until the 5bPreg peak aligned with the -26.16 IRMS.

Does this correctly represent the identification that is being claimed to be visually correct by Brenna, the Panel, and BB here?

thanks,
TBV

BustinBilly said...

There you have it. The three peaks are the same.

DBrower said...

OK. Remember what we did, we adjusted the horizontal scale to make those three peaks line up, so we think linear (proportional) adjustment is correct.

Now, look at figure 4, a zoom of the left hand side of figure 3a, where the "anchor" internal standard is supposed to be located.

There doesn't appear to be very good alignment with those peaks is there?

With the linear scaling we used to get the peaks in the middle and right to align, we don't have alignment on the peaks on the left.

How should we try to correct it, by changing the scaling factor, or shifting the left peaks into better alignment?

TBV


TBV

BustinBilly said...

There is nothing to correct. It is a qualitative issue.

DBrower said...

I think the point is clearer looking at zoom of Figure 3, which has less crap in the left edge. This will be figure 4a.

TBV

DBrower said...

By what do you mean "qualitative?"

Should the internal standard align in the GCMS and GC-IRMS the way the analytes of interest line up or not?

TBV

BustinBilly said...

It is qualitative. For identification only. Which is different from a quantitative issue, e. g. the T/E quantification that was negated. The identification of the three metabolites is quite obvious. What more is there to say?

docnzym said...

But the internal standards are suppose to be lined up or why would you call them an internal standard (other than for extraction efficiency purposes) By just looking at these three peaks and nothing else then it would seem that you are forcing the answer

BustinBilly said...

What does it matter if the quantified peak-internal standard difference is changed by 0.3-0.4 minutes? The patterns are the same. The peaks were identified in the GC/MS plot.

The plots are from different machines with corresponding columns. A difference is expected.

Unknown said...

bustinbilly,

I appreciate the time you have spent here, but are you really qualified to make this analysis? Having watched the testimony and read all of the documents, but not being a molecular chemist or experienced and analytical lab technician, I am not capable of drawing these conclusions. The three graphs could all show that he had an inordinate amount of coffee for all I know.

BustinBilly said...

Look, there are two pertinent points I haven't seen anybody argue away.
1.The substances were identified correctly in the F3 GC/MS plots.
2.The F3 GC/C/IRMS plots for both the A&B samples are nearly identical to the GC/MS plots. The three metabolites of interest are almost exactly identical.

The panel of judges was right when they decided this issue.

Eightzero said...

I am not at all clear that the judges decided this issue at all. It seems to me Botre explained it to them behind closed doors. He has credentials, but is hardly unbiased.

The arbitrators' duty was to evaluate the evidence, not to conduct scientific analysis. They do *not* have the credentials for that. BB if you do, this is important information. Are you a scientist, lab technician, or doctor? I am very interested to hear from a properly qualified, unbiased expert on this issue.

Unknown said...

Would it be possible to provide the units that go with the Y axis on both the GCMS and GCIRMS plots? If the units (rather than the measurements) are different (as I'm pretty sure they are) it might make it easier to note that in the GCMS plot we're measuring "apples" and there are several large peaks of apples that are associated with the steriods of interest. In the case of the GCIRMS plot we've also got several large peaks that happen to be similar in shape to the GCMS plot, but in this case we're measuring "oranges." A key point though is that in the GCMS we know which peaks represent steriods, but in the GCIRMS we have to figure that out from the retention times (or the shape of the peaks). If true then it might be easier to note that you can't assume that just because the big peaks of apples show up in about the same place as the big peaks of oranges that apples = oranges--essentially because the peaks represent different things (as noted by the different units of measure on the Y axis.)

DBrower said...

Regretably, I'm tied up with other obligations, and will return to this discussion perhaps much later tonight or tomorrow.

TBV

Unknown said...

One of the things that troubles me is that the operators were looking to make the two graphs appear qualitatively similar, and did that by massaging data and axes.

How easy is it to take something else with three peaks and cause it to line up, qualitatively?

Russ said...

The distortions of the time plots by non linearity have been discussed at some length but I am not sure the effect of non linearity on the quantity or vertical axis has been fully considered.

To attempt to use the cal-mix standards to establish the machines calibration and performance with the machines out of wack is treading on thin ice at best.

In addition, all of the experts who look at these plots and draw on either their experience or the published research, err in that all of the research should have been performed on in spec equipment. Further all of the experience (except, perhaps some from LNDD) anyone looking at these plots has should be from in spec, properly operated equipment.

How in the blooming lilies can you expect to say these numbers are convicting or for that matter any of the LNDD machine results can be compared to published research and then have valid conclusions made about them?

Any results from machines not being operated in spec should be thrown out, period.

I have wondered how LNDD could find itself in this state of affairs. One possible source of error that, I think, has not been much considered is that the gc/mass spec vendor that sold and maintains the machines for LNDD, may be a major contributor. The fact that the 'mouse ears' magnet lifting device has been left on one machine is probably evidence that the repair person assigned to maintain the machines has some issues.
The fact that no correct version manual for the machines was provided points back to this company as well.

Unknown said...

I'd like to revise my earlier "apples and oranges" post and use a slightly different analogy regarding the units of measure on the two diffent types of plots; height and weight of people (where each person is a peak and the labeled peaks are a type of steriod and a person whose name you know). The labeled peaks on the first GCMS plot represent the height of someone whose name you know while the peaks on the second GMIRMS plot represent the weight of people whose names you do not know. All you know is that the data collected (height and weight) on each person was done at the same relative time compared to their peers. The exercise is to figure out if the heaviest people in the second graph are any of the known (tall) people in the first graph. In most cases someone very tall who you know (the labled peaks on the GCMS plot) might also be very heavy, but not always. Without calculating relative retention times you don't know whether the peaks in the second GCIRMS "weight" plot belong to the taller people in the GCMS "height" plot, or to some shorter very heavy folks whose names you don't know. If you use relative retention times you might well discover that the big peaks on the GCIRMS chart really correspond to very small peaks on the GCMS chart, and vice versa, i.e., there are tall thin folks and short heavy ones.

wschart said...

As I recall, the testimony of the LNDD people was that they, in essence, eyeball things to ID the peaks of interest, while the testimony of the Landis experts was that you can't eyeball, but you have to go by the retention times. We also had the LNDD people admit that there were earlier runs which produced data they didn't like, and that they reran things, but erased the earlier data. We don't know what it was about the erased runs they didn't like, not do we know what, if anything, they did differently on the run they used. Could they have manipulated things in some way to produce results they wanted? Or is there enough variation is results naturally that repeated runs might eventually produce something that can be massaged to produce the desired results? Was there something in the earlier, erased data that was legitimately cause to question the data, or was their dislike of the data due to it not backing up the claim of doping?

The graphs may match up, but can they be trusted?

Unknown said...

russ,

I'm reading these comments with interest and for my own education.

I don't currently have anything to add on the science.

Regarding the "mouse ears", LNDD running their testing equipment out of spec, and lack of up to date manuals for certain testing equipment, there are a great many possibilitys.

It may just be poor training and management? It could be a lack of funds, or diverted funds? It could be indifference? It could be that LNDD has not paid their vendors for continued technical support. Thus they are doing the best they can without the latest info/updates that would be part of ongoing vendor support?

It could be any of those things, any combination of those things, or something else altogether.

Davis Straub said...

Billy has an interesting point, but he appears to be unwilling to elaborate enough on it for those of us who are not as familiar with this type of work as he apparently is (perhaps he learned it the University of Miami before he became a "studio flunkie") to proper assess it. I can't tell if he just is disdainful of the audience or whether he is not so sure of his position and that he is now unwilling to defend it. I would love to hear more from him.

DBrower said...

Hi, I'm back.

A warning to people to be nice, I'm trying to have a substantive discussion here.

I've moved the interesting stuff up above, and I'm responding to BB's 5:45 comment.

1. The GCMS identification isn't in question; that's not in dispute here.

2. This is what we are discussing, and we'll see if it gets argued away or not.

In the comment that Billy started with, he said "They are exactly the same".

We've shown with overlays they are not exactly the same, and Billy has changed to using "nearly identical" for the whole chart and saying the metabolites of interest are "exactly identical".

He says the lack if exact match on the low end near the SI is not relevant. We'll leave this aside for now.

I do want to offer two more versions of overlays for comparison.

Figure 5 is a version of the overlay where the SI is aligned, and the axis linearly expanded so the 5bP aligns. In this, we see a significant left shift of the 5bA and 5bP peaks.

Figure 6 is one where the overlay scaling of Figure 4a is left shifted so the SI seems to align. In this case, it doesn't appear anything on the middle or right aligns, so that might be pointless to pursue.

TBV

DBrower said...

Now a few more. Figure 7 is a zoom of the 5bP peak to show how it was aligned at the top. Figure 8 shows that the bottom alignment might be questioned, but the trailing peaks look OK.

Figure 9 is a zoom of what we think is the region of interest. In it, things are not quite as clear to me as they appear to be for BB. Here we see the peak identified as 5bA might align as well with a low GC-IRMS peak as with the one originally claimed, and the one labelled 5aA might align with the GC-IRMS peak that is ostensibly the 5bA.

Are the identifications of the GC-IRMS peaks still as obviously correct on visual examination as they seemed at first?

TBV

BustinBilly said...

Yes the identifications are as obviously correct as they seemed at first.

A good analogy would be different sized fonts. If you type the same sentence in the same font in Photoshop, the two sentences will overlay perfectly. Type the same sentence again in the same font-using a slightly larger point. A few letters will line up well on overlay, the rest will not. However it will still be the same sentence.

I hope that analogy helps. I don't see how quibbling over the overlay makes any difference.

Unknown said...

I sense another potential point of confusion. It's clear from all the evidence we're looking at that the steriods of interest are being processed by both the GC/MS and GC/IRMS systems. However, what we haven't talked about is WHAT ELSE may have been processed through both instruments. Since there are other peaks in both graphs, one can probably assume that other compounds are creating peaks in both graphs. That being the case, an interesting question is what compound is or could be producing the large peaks in the GC/IRMS graphs? One has to wonder if a steriod-like compound like the cortisone that Floyd was taking could show up as a small peak in the GC/MS (because it's not one of the compounds they're trying to detect), but have it show up as a large peak in the GC/IRMS because it has a high percentage of exogenous Carbon in it. This is pure speculation on my part, but if one believes that the large peaks in the GC/IRMS are not an indication of exogenous Testosterone AND they are unusual enough to have folks think that's what they indicate--then what might they actually be? One of the clear differences in Floyd's biochemistry from every other tour rider in 2006 was his exempted use of Cortisone to treat his hip. I just wonder if that could be what's confusing the GC/IRMS peaks. Note, this is pure speculation, but a knowledgeable biochemist might be able to cast some light on the idea.

Unknown said...

With regards to Bustin Billy's comments about the overlays... I'd like to pose a question to Bustin Billy. How much horizontal difference in the overlays would there have to be for you to believe that the peaks don't match--5%, 10%, 20%, 50%, and how much difference in the vertical heights? I believe you may be allowing yourself to be deceived by the relative height and relative horizontal separation between two sets of peaks that may or may not be related. Just because their vertical shape is similar doesn't mean anything when what's being measured vertically are completely different characteristics (height versus weight, for example) and the horizontal positioning only looks similar when scaled--as TBV has pointed out they don't line up very well, certainly not within 0.2 minutes or 1%.

DBrower said...

Please be patient while I try to figure out what is obvious to you, and not yet to me.

In Figure 9, we can see that innaccurate horizontal alignment can result in peaks being completely misidentified. The GCMS peak for 5bA is closer to a small GC-IRMS peak just short of 15:00 minutes that has not had it's CIR measured. The GCMS peak identified as the 5aA peak nearly sits on top of the GC-IRMS peak that is measured to be the 5bA peak.

How is that potential misidentification not a problem?

If that misidentification can't happen, what criteria are you using to decide which peak in that region aligns with which peak in the other chromatogram?

TBV

Unknown said...

TBV,

I have been following this for a while and am glad to see the discussion with BB. In Figure 6 I see that the peaks to the right do not line up. If I understand correctly the scaling for the Figure 6 overlays was developed from that of plots in Figure 4a. I am suspicious that the scaling developed from 4a will work on the plots in Figure 6 given the way it (the scale) was developed. I think that you had to work with images either scanned or copied from the originals. If there are differences in the magnification of the images from one reproduced plot to the other the scaling for one pair of overlay plots might not work for another pair. This is where having the data files is important, especially if you need to select between peaks located closely together. If you had the data files it would be easy to develop a function (RRT?) that would locate the peak on the GC/IRMS based on data from the GC/MS. Without the data you are stuck with trying to resolve the peaks with the graphical method that you are using.

If I am off base with the assumption that the scale for the Figure 6 plots was developed from the 4a plots then please forgive me and move on.
Regards,
Rich

Larry said...

TBV, I'm enjoying this exercise in "Fun With Photoshop". But aren't you trying to do with pictures what Dr. Augenstein was trying to do with the numbers? Aren't you just doing a pictorial demonstration of relative retention times? What the FL majority decision ruled was impossible? And didn't Dr. Augenstein's numbers show that the peaks cannot be made to match up mathematically?

As between a picture show and a numeric analysis, I think I'd go with the numeric analysis ...

DBrower said...

BB said look at the graphs so I'm looking at the graphs.

Larry said...

TBV, yes, but why? The graphs are just pictorial representations of the numbers, right? So you can line up the pictures, and I admit, it's an interesting exercise, but at the end of the day, aren't the numbers more precise?

And there's that other question, where the majority of the arbitrators ruled that doing relative retention times is not a valid way of doing the testing. Aren't you just doing relative retention times, with pictures instead of the math?

But then, when the majority says that you're supposed to eyeball the graphs to identify the peaks, aren't they just advocating some kind of fuzzy relative retention time method?

It may sound like I'm asking questions when I already know the answers, but I don't really know the answers here. I swear. I keep reading and reading this stuff, and I don't seem to get it.

DBrower said...

I suspect the gravitas is that there is a visual gestalt that happens when you see the plots and conclude they are isomorphic. Therefore, looking is a different way of interpreting the same data.

I'm attempting to identify aspects of the visualization that seem to encourage that conclusion.

TBV

Larry said...

TBV, I appreciate what you're trying to do. Really. You're trying to get someone here to explain to us (even defend for us) what the majority of arbitrators were trying to say in their opinion. I sincerely appreciate this effort! I think that our first job is to try our best to understand the majority decision on its own terms, and not simply to bash the decision because it did not come out the way we wanted it to come out.

But are you really telling me that the FL case boils down to the concept of "visual gestalt"? Wow. I'd smile if it wasn't so sad.

I don't think that the concept of "visual gestalt" is good science, though I'm not a scientist. I AM a lawyer, and I don't think that "visual gestalt" would hold up in a legal proceeding.

Now, if what we're talking about is isomorphism, then I guess we have something that at least is a math concept. For those lurking who, like me, might not know the meaning of "isomorphism", I'll quote the following from Wikipedia:

"The word 'isomorphism' applies when two complex structures can be mapped onto each other, in such a way that to each part of one structure there is a corresponding part in the other structure, where 'corresponding' means that the two parts play similar roles in their respective structures."

Wikipedia suggests a few examples of isomorphic structures: a deck of 52 playing cards with green backs is structurally isomorphic to (with?) a deck of 52 playing cards with brown backs. The playing cards are isomorphic because if we want to play cards, it doesn't matter which deck we use. Another example: the clock tower in London (Big Ben) is isomorphic with a wrist watch, since both use isomorphic mechanisms for reckoning time.

Now ... you're not seriously suggesting that ANYONE thinks that you can base an adverse analytic finding on whether some lab technician thinks he (or she) detects an isomorphism between two graphs overlaid one on top of the other? I mean, on a certain level, aren't ANY two peaks isomorphic? To paraphrase Monty Python's description of Mt. Kilimanjaro, they're all a tricky climb to the top, at which point they tend to slope away rather sharply.

I'll still leave out there for comment whether BB's comparison of pictures is just an imprecise way of comparing relative retention times, and if this is so, how such a comparison could be valid given the majority opinion's rejection of relative retention time analysis.

Duckstrap said...

Actually, Billy is doing is exactly what the arbs did, and apparently what the technicians did, and that is to identify the peaks via a "visual gestalt". The supposed purpose of the internal standard was to "anchor the retention times", but as the stretching excercize shows, it can't actually serve as an anchor. And yes, the problem is precisely that "ANY two peaks are isomorphic", but since the relative retention times aren't the same, you really have no idea what is actually in the IRMS peaks--assuming for a minute that retention time in some form or fashion is an appropriate means of identification. What we do know from the IRMS chromatogram is that there is a source of CO2 there, but what it is remains a mystery. So the lab techs and the arbs are back to Bustin Billy's "looks alot like the other one".

Larry said...

Kevin, I'm experiencing great confusion here. I'd understood that mass spectroscopy was more complicated than, say, a pregnancy test. But I also thought that it was pretty well accpeted science. In other words, it might be complicated, but it's reliable if performed correctly, and there's not a lot of controversy in the scientific community about how to properly perform or interpret a mass spectroscopy test.

Now, it looks like any of the following might be the case: (1) there's no controvery about how to perform a mass spectroscopy analysis, but there's some "art" in the process -- it requires some skill and judgment to analyze the results of a mass spectroscopy test, and reasonable people can interpret the results differently; (2) there IS a legitimate controversy in the scientific community on how to correctly interpret the results of a mass spectroscopy, and the FL case falls squarely in the middle of this controversy -- the prosecution has adopted one side in the controversy and the defense has taken the other side; (3) there is NO controversy, and there's also NO art in the interpretation of the results of a mass spectroscopy -- there's an objective and reliable method used to conclusively interpret these results, but only one side of the FL case is using this method.

So, in the multiple choice test I've posed above, is the answer (1), (2) or (3)? And please don't tell me that the correct answer is "none of the above"!

m said...

TBV

Key assumption proportionality?

"OK. Remember what we did, we adjusted the horizontal scale to make those three peaks line up, so we think linear (proportional) adjustment is correct."

Why do you think you can do this and get an exact line up?

According to Brenna you can't compare the retention times between the two machines.

"Dr. Brenna’s testimony specified that it could still be possible that the retention
times might be proportional to one another (and within TD2003IDCR), but it
won’t always work out in that way."

So you won't necessarily get an exact lineup after proportional adjustment.

Rather they look to see if the peaks are same (in the same position and of the same magnitude?) from the identified reference point.


Brenna:
"Instead, the lab compares the peaks and the
sequence of the peaks from the GC/MS and GC/C/IRMS to identify the
metabolites and the endogenous reference compounds. Specifically, to identify
the substances in question, one would compare the pattern of peak heights and
retention times in the GC/C/IRMS chromatograms, anchored by the internal
standard with a known RT, with the pattern of peaks heights and RTs in the
GC/MS chromatograms obtained from the same aliquot of the sample."

When I look at the two plots separately I can see why Billy says they have the same pattern. I can't tell what you are getting at with your overlays. Visually they are too messy.

Pesty said...

m,
I think we are going down this path because Billy's original post said that Floyd was guilty because the plots were "exactly the same".

Forgetting about retention times as instructed, I think what TBV is trying to show with the overlays is that they are not exactly the same.

When you scale the plots to get the middle and last peaks to match up, the SI does not, and when you get the SI matched up, the other peaks are off.

Based on Billy's argument as I understood it from the beginning; if you compare the graphs, you should be able to get the peaks to match up exactly when you adjust the scale linearly. TBV has shown with the overlays that this is not the case. There is no way to get the SI, and the other two peaks to match up, and they are off by a significant amount.

I could be off on this one, but that’s the way I’m reading it.

Pesty said...

Just to clarify what I think we are doing here… Looking back at the majority decision, paragraph 186:

186. Dr. Brenna’s testimony specified that it could still be possible that the retention times might be proportional to one another (and within TD2003IDCR), but it won’t always work out in that way. Instead, the lab compares the peaks and the sequence of the peaks from the GC/MS and GC/C/IRMS to identify the metabolites and the endogenous reference compounds. Specifically, to identify the substances in question, one would compare the pattern of peak heights and retention times in the GC/C/IRMS chromatograms, anchored by the internal standard with a known RT, with the pattern of peaks heights and RTs in the GC/MS chromatograms obtained from the same aliquot of the sample.

It looks like this is what Billy is looking at and asking us to compare. The issue is, that when you get the peaks to line up, the SI, which should be the anchor, does not. When you match up the SI the peaks do not match up. If the peaks do not match, you cannot reliably say what metabolites you are looking at in the IRMS.

Someone please correct me if I’m wrong on this.

Duckstrap said...

To Larry,
When performed well, the answer is more like 3, no controversy and not a lot of art required. If the chromatography were performed identically between the two machines, which it should have been, but was not, and the MS identification were performed correctly, as it should have been, but was not, then either Floyd would have been definitively exonerated or definitively convicted, which, in my book, he was not. The arbs' ruling makes clear that they did not understand the requirements for the analytical technique, and they pushed aside many years of settled separation science in order to reach that verdict.

m said...

Re: linearity

I don't read Billy's comments to demand or suggest linearity or proportionality.

In any case I don't read the decision to suggest that.

Indeed, the majority decision in their mathematical example seems to suggest that what TBV is doing can never result in a match.

They say the retention times of the IRMS GC take longer than the GC-MS machine by some constant amount of time. They use the example of a relative retention time ratio of 10 to 5, becoming 11 to 6, after you add a 1 minute constant.

Well if in fact the IRMS retention times are off by some ADDITIVE constant relative to the GCMS machine, then you can't correct for that by MULTIPLYING in some proportionality correction. By stretching the scale I assume TBV is doing some multiplying. At least that is how I understand the example and the math.

DBrower said...

m,

yes Brenna did at the end, testify to the "visual gestalt" method, which is what we're exploring in this discussion.

This was after he earlier testified that relative retention time was the method, and the technicians testified relative retention time was the method, and USADA's brief said relative retention time was the method.

This contradiction seems not to have bothered the Majority, as they did not comment on it, but we'll leave that aside for another post.

We will discuss whether relative retention time was properly dismissed by the Majority, in a different post. Hint: it deals with the non-linearity, when done correctly.

Right now, I'm going to continue on the validity of "visual gestalt" by looking at more graphs.

TBV

m said...

Re: visual gestalt

Looking again at the graphs, I have to agree with Billy that for F3, the GCMS and the GC-IRMS are a pattern match.

You can pretty much count the "peaks" in each graph and they match up, although it's a bit hard to distinguish the GC-IRMS printout because it's less clear on the far left. Nevertheless, there seems to be the same number of peaks and in the same sequence in both graphs.

So if you take the 4 peaks in the middle, the middle two higher peaks of which are the 5b Andro and 5a Andro respectively, they match up in both the GCMS and GC-IRMS. I don't think there could be much doubt that the 5b in the GCMS corresponds to the 5b in the GC-IRMS, even if one might be displaced on the horizontal scale. The pattern match is simply staring you in the face.

As I understand it they know for sure that there must be a 5b Andro in the sample, the only problem is to locate it. It's pretty obvious that the second of the four middle peaks is the 5b Andro and not any of the other 4 middle peaks.

BustinBilly said...

TBV,
USADA's pre-trial response brief states the technicians used RRT for GC/MS identification not inter-instrument comparison.

DBrower said...

OK, to throw a little more visual example around, Fig 11 is the GCMS of the the A sample F2, and Fig 12 is the IRMS.

This fraction isn't in dispute.

Figure 13 is an overlay of the previous two that lines up the SI and the rightmost labeled peak. The other labeled peaks seem to line up with TD2003IDCR tolerances, so this should make everybody happy.

What's interesting is the larger GCMS peaks the follow, circled in Fig 13. They were very big in the GCMS, but almost become noise in the IRMS. This seems to say that peak height may not be as good a marker as is assumed in by the visual gestalt.

TBV

DBrower said...

Hi, BB...

Can you find a cite in the USADA brief where they say how they do the matching between the IRMS and the GCMS? It seemed to me the only discussion of identification in anything involved RRTs.


M,

The panel got their "math" example wrong. If they'd figured it out correctly, the example they gave as "broken" worked just fine. You are supposed to subtract the fixed offset before you do the multiplcation, so (11-1) / (6-1) works out to 10/5 just like they said it didn't.

Visually, this is the same as aligning the left peaks (correcting the offset by addition or subtraction) and then scaling (doing the proportional multiplication).

TBV

m said...

TBV,

What is the authority for these two statements:

1. "You are supposed to subtract the fixed offset before you do the multiplcation"

2 "and then scaling (doing the proportional multiplication)." (I asked before, on what basis are you claiming you are entitled to do proportional scaling?

Duckstrap said...

m--
You can google "relative retention time chromatography", and the first link you find (http://www.iupac.org/publications/analytical_compendium/Cha09sec237.pdf) will take you to a pdf document entitled "9.2.3.7 Retention Parameters in Column Chromatography". It has a section (p.2) describing the Adjusted Retention Volume (Time). It then goes on to discuss Relative Retention (r) on p. 4. For those who are interested, IUPAC is International Union of Pure and Applied Chemistry. This is pretty standard stuff.

m said...

Kevin,

4 questions:

1. are you a scientist?

2. does your reference purport to answer my two questions to TBV.

3. If it's "straightforward stuff", can you explain this on page 4 of your reference:

"The ratio of the adjusted or net retention volume (time) or retention factor of a component relative to that of a standard, obtained under identical conditions:
r = VRi'/VR(st)' = VNi /VN(st) = tRi'/tR(st)' = ki /kst. Depending on the relative position of the peak corresponding to the standard compound
in the chromatogram, the value of r may be smaller, larger or identical to unity."

4. Does this guidance apply to GS between two different machines as in this case, GCMS and GC-IRMS.

And if I'm not running afoul of Bill Hue's standards of decorum:

Are you really trying to be helpful?

m said...

In 4 above:

GS should be GC gas chromatography

m said...

TBV

RE: your 11, 12, 13.

One man's "noise" is another's peaks.

The peaks all match up. It's true that the peaks you point to in 12, the GS-IRMS seem proportionately lower than the corresponding peaks in 11, the GSMS, but they are identifiable peaks in the proper sequence. Again this shows that the substances are properly identified for the IRMS analysis.

Plus the peaks you've identified were of substances of no interest to the analysis.

wschart said...

M

Kevin said that it's "standard stuff" not "straight forward". This means that there is a standard way of doing this as accepted by the IUPAC, not that it is "straight forward", i.e., easy to understand.

The equation you quote is saying that r, the ratio of the adjusted or net retention factor of a substance compared to a standard can be calculated from any of the ratios of whatever those VRs, VNs, tRs, and Ks stand for. The i indicates that the data point in question came from the sample in question, and the st that it came from the standard.

What this seems to be saying is that it is possible, indeed the standard method, to mathematically adjust the RTs to make the ID, and not to use some "visual gestalt".

m said...

wschart,

Are you a scientist?

What is your basis for this statement:

"it is possible, indeed the standard method, to mathematically adjust the RTs to make the ID, and not to use some "visual gestalt""

I hope you are not claiming that that one equation is the basis for your statement. Especially, since you're not even sure what all the terms mean.

Duckstrap said...

To m,
To your answer your questions (maybe out of order):
4) Yes, I am trying to be helpful.

1) Yes, I am a scientist--I have contributed in the past under the name Duckstrap on DPF. That handle is taken on google (probably by my brother), so I can't use it here.

2) Yes, the reference does address your questions, though on rereading the passages, the notation may not be as clear as I thought.

3) The term "adjusted retention time" refers to the measured retention time minus a constant time to account for the "plumbing time". Unadjusted retention time refers to the total time from injection to peak, including plumbing and chromatographic separation. The [adjusted--my addition] relative retention, as defined on p. 4 of the reference is the ratio of the adjusted retention times of the internal standard and the target. Under consistent chromatographic conditions, this would be portable between systems. The LNDD systems did not use consistent chromatographic conditions, therefore the separations are not comparable.

Unknown said...

TBV,
I work in the R&D lab of a large consumer products company that
develops personal care products. I'm not an Analytical Chemist, but we have an analytical chemistry department that has several GC/MS units (but no IRMS). I sent the head guy this thread and then asked him if he would expect the peaks to line up or scale. He thought they probably would because the same GC column was used, so the metabolytes would elute at the same time. There would either be a constant offset
for the difference in time between that required for MS (in the GC/MS) and the combustion process (in the GC/IRMS), or a scale factor. If, for some reason, this was not the case, the only way he thought you could overcome it would be the run standards that were purified forms of the metabolytes themselves (on the GC/IRMS), because you absolutely
have to know the time they should appear to be able to identify them. In other words, you cannot identify metabolytes by simply eyeballing or matching up the output from the two machines, because the whole principal that they work upon is elution time. He thought using something other than time to identify metabolytes was speculation, especially because urine is not a clean matrix.

Hope this was helpful.

m said...

Kevin,

thanks, that was a little more helpful.

Clarification re this statement:

"The [adjusted--my addition] relative retention, as defined on p. 4 of the reference is the ratio of the adjusted retention times of the internal standard and the target. Under consistent chromatographic conditions, this would be portable between systems. The LNDD systems did not use consistent chromatographic conditions, therefore the separations are not comparable."

1. Since the two systems (GCMS & GC-IRMS) did not use consistent chromatographic conditions one would not necessarily expect their respective retention/rrt/arrt times to match, and TBV would not be entitled to proportionately adjust the graphs and expect an exact match. Yes or no?

2. What are "consistent chromatographic conditions" that were not met in this case? For example, page 1 mentions recorder speed and gas flow rate as variables.

3. I've read that both the GCMS and GC-IRMS were tested against an internal reference standard and the retention times for each were within the technical standard of 1%. I.e. that each machines was accurate. Why shouldn't that be enough? Why should the rrt's between the machines have to match withing 1% as the Landis team claims?

4. Looking at TBV charts 1a and 2, is there any doubt that 5b and 5a Andro are properly identified for the IRMS?

many thanks

Duckstrap said...

m--
Regarding your questions:
1) The LNDD lab techs, Brenna in his initial testimony (but not in his rebuttal), and any number of other documents (I think TBV could give you chapter and verse on which ones) asserted that they had used relative retentions times to identify the different compounds--i.e. stretch the chromatograms as TBV has done. What TBV has shown is that stretching chromatograms does not lead to peaks in the expected places. So what are the peaks that are measured in the IRMS? I don't know, do you?

2) There are a number of inconsistencies, specifically, the temperature profiles are substantially different.

3) The rrt's should match first because that is how you would have some assurance that you were measuring what you thought you were. Secondly, the TD2003IDCR requires them to be consistent, and the ISL has a requirement that the compounds be unequivocally identified. That was not met in this case, in spite of what the arbitrators stated.

4) Obviously, I think there is a substantial possibility of misidentification, and what is more, nobody has any idea what else is in those peaks, because the mass spectral data were never properly characterised.

wschart said...

M:

No i am not a scientist.

You left something out of your quote of my post. I'll paste the full statement:

What this seems to be saying is that it is possible, indeed the standard method, to mathematically adjust the RTs to make the ID, and not to use some "visual gestalt".

I am not trying to make a definite statement that it is possible , but that the statement containing the equation seems to be saying that. I am enough of a mathematician to know how to interprete a simple equation like this, even if I don't know what all the variables represent.

m said...

Ken,

As to your answers to:

Q1. You didn't seem to answer this question. If the chrmomatographic conditions didn't match, you can't necessarily expect the relative retention times in the GCMS and the GC-IRMS to match exactly. Yes or no? Thus you can't expect an exact match by proportional stretching. Yes or no? As quoted: Brenna said you cannot expect an exact match of relative retention times. In fact the matching is very close, but not exact.

Q4: How can you think there is a misidentification in graphs 1a and 2? There are four peaks in each. They all have the same shape. The four peaks in one have to map onto the four peaks of the other. If, as I recall TBV did, you try to shift the peaks over and map the big 3rd peak (5a Andro) in the GCMS onto the small 4th peak of the GC-IRMS, then you have the 4th peak of the GCMS mapping onto nothing. It doesn't make any sense.

Davis Straub said...

Don't you mean Kevin, "m"?

m said...

Davis,

thanks, yes I do mean Kevin.

And in Q4, TBV may have shifted the peaks in the GC-IRMS onto the GCMS, rather than vice versa as I originally stated, but the point is the same, you have a peak mapped onto nothing.

Unknown said...

Hey M, you seem to be attacking the posters rather than their arguments.

Are you a lawyer, by chance?

Unknown said...

Kevin,
From your 11:30 post yesterday morning you said:
The LNDD systems did not use consistent chromatographic conditions, therefore the separations are not comparable."
Are you saying they didn't use the same column on both GCs?

From statements one and four in your post, I take it that you think Floyd got hosed (that's where I'm landing after talking to the Analytical Chem guys at work)?

I'm assuming that you have an analytical chem background and have worked with GC/MS. If so, is my understanding correct that you really can't identify a peak unless you know ahead of time what time it should appear?

Regards