GCMS and HPLC -- any relationship?

[readalot]

There it is. Parity plot for "MCT model" and "other oils" model. All collapse onto master line with slope =1 and practically zero intercept. RMSE around 2%.

MCT model is solid with 8 data points. Basically order GCMS and get HPLC measured number as model output.

Thank you @Photon and @TallandSmall!


Model equations and ANOVA results to follow.

@MuscularMD

@janoshik
@Millard

What a community!

 

[janoshik]

I'm not going to pretend I understand what's up in my sleep deprived state, but I'm amazed nevertheless!

 

[readalot]

I'm not going to pretend I understand what's up in my sleep deprived state, but I'm amazed nevertheless!

Let's talk when you have time. The math is cool, models are cool, but want to be respectful as this affects your business and I don't want these models used incorrectly.

The other fun part about this analysis is a really low purity raw will show up as an outlier in the model.

So you have here a vendor raw purity BS detector just by running GCMS on their injectable product. And you get very accurate estimate of what the HPLC result would be on the injectable.

 

[Photon]

Let's talk when you have time. The math is cool, models are cool, but want to be respectful as this affects your business and I don't want these models used incorrectly.

The other fun part about this analysis is a really low purity raw will show up as an outlier in the model.

So you have here a vendor raw purity BS detector just by running GCMS on their injectable product. And you get very accurate estimate of what the HPLC result would be on the injectable.

We've only got n=8, I'd think we should try and get more samples first before jumping to conclusions lol

Majority won't do GCMS, since vendors won't reimburse both GCMS+HPLC.
However if we can get more people to do GCMS instead, i think it's better business wise since its priced higher, and also it helps improve the quality of the oils.

All GCMS in that list, was only performed in 2025.
And from those GCMS, we spotted vendors with (1) contamination, (2) lieing about oils used and also using (3) EO / possibly reselling china gear.

 

[readalot]

We've only got n=8, I'd think we should try and get more samples first before jumping to conclusions lol

Respectfully, I understand your concern. However thats the whole point of statistics. The ANOVA results (not shown yet) on n=8 are extremely compelling. And I am not just talking about the correlation coefficient. I am also referring to the F-value as well as the p values on the regression slope and intercept. Both of those are very SIGNIFICANT. As well as the RMSE on the overall model. So with n=8 I can make some very strong conclusions at this point.

Would it be nice to have more data? Sure. However, the model as is across multiple vendors, multiple compounds is still astounding IMO.

Majority won't do GCMS, since vendors won't reimburse both GCMS+HPLC.
However if we can get more people to do GCMS instead, i think it's better business wise since its priced higher, and also it helps improve the quality of the oils.

We are working on that. It is improving. And with the combo price discount it will be even better.

All GCMS in that list, was only performed in 2025.
And from those GCMS, we spotted vendors with (1) contamination, (2) lieing about oils used and also using (3) EO / possibly reselling china gear.

Wonderful!

 

[readalot]

I'm not going to pretend I understand what's up in my sleep deprived state, but I'm amazed nevertheless!

You sound so tired all the time. I am sorry Brother but you are hard worker.

Sometimes I think I need to stop adding to your workload. But that would not be right. Humans dont want to be happy. They yearn for purpose. To make a difference. To do work that matters. So congrats. You are a rich man!

 

[readalot]

Any interest in me sharing the remainder of the details? Stats, model parameters?

Also thinking about getting more HPLC+GCMS pairs to further validate the MCT model and add training data for GSO model.

Two models. Changes the game.

When model ready...sources do HPLC testing, members do GCMS. Everyone wins.

Thank you @Photon!

 

[newguy2]

Interesting stuff here. I have no clue what the chart means lol.

 

[readalot]

Interesting stuff here. I have no clue what the chart means lol.

Model predictions vs Jano HPLC test results. Called a "Parity Plot" used to gauge model fitness.

Dead nuts.

For MCT based gear, we can calculate what gear will show by HPLC (mg/ml) by using GCMS data (API %). Within ~2%.

Need more GSO data and should have that dialed in as well.

 

[Photon]

Model predictions vs Jano HPLC test results. Called a "Parity Plot" used to gauge model fitness.

Dead nuts.

For MCT based gear, we can calculate what gear will show by HPLC (mg/ml) by using GCMS data (API %). Within ~2%.

Need more GSO data and should have that dialed in as well.

Likelihood of anyone doing GCMS on non-MCT is probably 0%.
Too little vendors sell it.
I wont bother with anything except MCT lol

 

[readalot]

Likelihood of anyone doing GCMS on non-MCT is probably 0%.
Too little vendors sell it.
I wont bother with anything except MCT lol

Plenty of vendors have GSO. Humor me.

 

[Photon]

Plenty of vendors have GSO. Humor me.

Only ones selling it are US Dom.
Little as in relative to amount of volume being sold. Till date I've never seen a GCMS done on GSO ugl..

 

[readalot]

Only ones selling it are US Dom.
Little as in relative to amount of volume being sold. Till date I've never seen a GCMS done on GSO ugl..

Is that a dare?
Hypothesis: We will get support from quality DOM sources.

 

[Photon]

Is that a dare?

Sure, if it means we'll get more data

 

[readalot]

Sure, if it means we'll get more data

 

[Photon]

Sure, if it means we'll get more data

Any idea if there's anything from SST? I don't really use that board.

 

[readalot]

Any idea if there's anything from SST? I don't really use that board.

None.

 

[readalot]

Any idea if there's anything from SST? I don't really use that board.

I think I may know where to get some more data.

 

[readalot]

Looking over the general approach a little more, makes perfect sense to have one linear model for MCT and another for GSO. Vapor pressures, etc.

Should cover just about everything. Inclusion of BA on the GC peaks should not change anything really in terms of model fit. 1%?

 

[janoshik]

You sound so tired all the time. I am sorry Brother but you are hard worker.

Sometimes I think I need to stop adding to your workload. But that would not be right. Humans dont want to be happy. They yearn for purpose. To make a difference. To do work that matters. So congrats. You are a rich man!

Oh, I wouldn't want it any other way!

Thank you and cheers