https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

IASIM 2018 Advertisement

Web based (Online) Calibration Service Provider

bikingisfun's picture
Forums: 

Hello,

Presently the trend in using NIR in the Feed and Animal Industry seems to be going "online" wherein the prediction equations (calibrations) are stored in a web server. You basically scan your sample in your own NIR and upload the spectra in the service provider web platform. Results will be predicted and stored on the web within several minutes.

The special features of this platform is the "real" nutrients are being quantified in the test so they highlight "in vivo" test such as digestible amino acids (DAA) and Apparent Metabolizable Energy (AMEn), while some offers phytate phosphorus to advertise their services as compared to the proximate test which Nutritionist don't often use.

Since these test are quite expensive, very exhaustive, definitely non routine. How would you know that the calibration itself or the prediction it provides is true and accurate?

The variables they (providers) want to control on the Users end to reduce variability are particle size (sample ground to 1mm and below), same NIR model, standardization of equipment on certain models and adhere to NIR manufacturers usage instructions. Is it enough to follow their recommendations to ensure accuracy. How about on their end, what are the measures or criteria to look for to verify that what they say on their platform will reflect on real samples.

Thanks.

Kim

 

 

jcg2000's picture

Kim:

You raised very good questions and the answers can only be got from your service provider.

I am not in food and feed industry, but I have heard many cases where one model developed in a central lab is deployed remotely to many statelite NIR instruments. Obviously instrument standardzation and calibration transffer are critical for prediction accuracy. Someone from agriculturial industry may have better insight into this practice.

In your case, do you peridically scan quality control samples and upload the spectra to the web and see if the web server can accurately predict these quality control samples? If not, you certainly have very good reason to suspect the quality of service you subscribed.

 

Jerry Jin

hlmark's picture

Kim - I agree with Jerry, that you should ask the service provider how they do what they claim to do? Also that regularly running QC samples is a good idea in any case.

Beyond that, you might want to consider "spiking" a sample with a known small amount of analyte. This is tricky because often adding pure analyte to a sample will change its optical properties and generate errors. A related action would be to take two samples, with known "low" and "high" values of the analyte and make two or three intermediate mixtures from them. The you can see how well the service can measure those samples. Ideally the "base" (everything in the sample except the analye) should be the same in the "low" and 'high" samples.

Good luck

\o/

/_\

 

bikingisfun's picture

Jerry and hlmark:
Thanks, you are right to say that the answer to my question will be best explained by the provider themselves. I posted my concern here because I guessed I am a bit confused on how to treat a calibration equation which you don’t have control (you don’t own/created yourself) but It seems the treatment should be no different from the usual validation and maintenance check as YOU and HLMARK implied on your response to my query.
To give you an overview: I’ve had several discussions with these providers and I think they were able to get more than enough sample of almost all agricultural commodities important to animal feed industry from various region “globally” and put them in their data set. They were able to do that because their company, which supplies feed additives, is all over the world and the samples from their client are shipped back to their central lab – so they have the means to collect representative samples globally.
The question is will the calibration they created works on our sample locally or on materials we import.
Of course they said YES. They did discuss their measures to ensure the prediction accuracy of their calibration service and its limitations and those are as follows:
1. Overview of calibration development ; others includes a description of their in vivo experiments
2. ISO Reference Laboratories
3. Statistical Summary of their Calibration which includes n, SEC, RSQ, SECV, RPD etc.
4. Standardization of equipment, sample preparation, environmental condition etc
5. Most calibration will expire which will only work again when you renew your contract except the other provider which have a service on per sample basis – you pay what you use.
6. Validation using locally sourced samples plus monthly Proficiency Test for the pay as you go service provider and Regular Annual Ring Test for the Feed Additive Companies with occasional random verification assessment..
7. There is a limit on the samples you can upload for the digestible parameters: for example let’s say a maximum of 3 or 10 samples a month. So if you are receiving 70 rails of corn or wheat per month, you are limited in scanning and uploading only 3 that represent those 70
With proximate, I think I’ll be able to verify their claims because the results can be verified by local wet laboratories. The problem is the Total Amino Acids which I don’t know if anyone does that except perhaps research lab in EU? A bigger challenge is how you will assess the “in vivo” experiments such as digestible nutrients such DAA, AMEn as well as phytate phosphorus which is the unavailable portion of phosphorus in plant materials.
Since we don’t have the capability to do those, I guess the most practical way to check them is to follow the basics and hlmark qc samples blending of high and lows if ever I come across them and use them wisely in checking the “digestible/invivo’’calibration equation.

Thank you for your response. Appreciate it a lot.

Kim

<p>kim</p>