| Author | Message | ||
     kerry walsh (walshk) Junior Member Username: walshk Post Number: 9 Registered: 2-2007 |
Hi Rui, Howard (at the tisk of continuing this 'fruit' conversation in 'complex solution' thread) Howard, fortunately temperature compensated refractometers are pretty standard (generally compensated to 20oC). Rui, yes, the extent of H bonding changes wih temperature, and thus the position of the absorption features - a bugbear for anyone dealing with high moisture content samples. Incorporating temeperature variation in the model is a tried and true solution, though at the loss of some precision. I add some references that you might find useful (eg the first one deals with temperature) Guthrie, J.A., Reid, D., and Walsh, K.B. (2005b) Assessment of internal quality attributes of mandarin fruit: 2. NIR calibration model robustness. Australian Journal of Agricultural Research 56, 417-426. 58. Guthrie, J.A., Reid, D., and Walsh, K.B. (2005a) Assessment of internal quality attributes of mandarin fruit: 1. NIR calibration model development. Australian Journal of Agricultural Research 56, 405-416. 80. Subedi, P. and Walsh, K.B. (2011) Assessment of sugar and starch in intact banana and mango fruit by SWNIR spectroscopy. Postharvest Biology and Technology 63, 238-245. DOI: 10.1016/j.postharvbio.2011.06.014 Subedi, P., Walsh, K.B., Owens, G. (2007) Prediction of mango eating quality at harvest. Postharvest Biology and Technology 43(3), 326-334 | ||
| Rui Manuel Farinha das Neves Guerra (rmfguerra) New member Username: rmfguerra Post Number: 5 Registered: 6-2011 |
There is a paper by Bart Nicolai and co-workers about the problem of temperature in Brix model calibration. Their conclusion was that the best approach is to mix measures made at different temperatures in the same calibration model rather than produce a calibration model for each temperature range. The basic problem is that if you calibrate a model at ambient temperature and then apply it to fruits at temperatures much lower (or higher), then the model gives wrong predictions. And I think this happens because of the dependence of the water absorption bands in temperature. Rui | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 483 Registered: 9-2001 |
For some reason I'm not seeing the full thread when I clicked on the link in the messages by Dave Hopkins or Kerry Walsh, about the relation between Brix and Refractive Index. I'll have to simply assume that they're still there. In any case, I wanted to comment that my recollection about R.I. is that it's exquisitely sensitive to temperature as well as the composition. Therefore both the NIR and even internal reflection measurements are susceptible to that error source. \o/ /_\ | ||
| Rui Manuel Farinha das Neves Guerra (rmfguerra) New member Username: rmfguerra Post Number: 4 Registered: 6-2011 |
Dear All, Thanks for all your suggestions. You are really kind in sharing your knowledge. Just a minor add relatively to the last two posts. Working with mango I have observed what Kerry described: the Brix reading was continuously changing on a digital refractometer. I attributed this to particle precipitation within the drop deposited onto the instrument's cup. But the traditional ocular refractometer was immune to this effect. On the other side, the effct was not observed for 'Rocha' pear in any of the instruments. Since then I prefer to use the ocular refractometer, for two reasons: a) it's more immune to particle setting; b) it's the instrument used in real life conditions, in packing houses and factories, and these are the conditions that you want to re-create. Thanks again, Rui | ||
| kerry walsh (walshk) Junior Member Username: walshk Post Number: 8 Registered: 2-2007 |
Hi Dave Brix is indeed a measurement of index of refraction of the water solution. Now the slope of RI to concentration differs for different solutes, so there is a bit of an issue as fruit sugar composition changes (eg during maturation the G:F:Suc ratio) But your question is about suspended solids... the digital refractometers measure the angle of total internal reflection, and are pretty insensitive to suspended solids (the internal reflection interacting with only the first few micrometers of solution in context with the [sapphire] crystal face) However, the measurement is sensitive to particles settling out onto the crystal - so the trap for young players is with fruit that contain starch - the granules are expressed in teh juice and are heavy enough to settle...so repeated measures of the one sample will give different values...you could centrifuge etc, but in practice the common approach is to mix well and take the first measurement Kerry | ||
| David W. Hopkins (dhopkins) Senior Member Username: dhopkins Post Number: 217 Registered: 10-2002 |
Hi all, This has been very interesting. I have a basic question. My understanding is that Brix is a measurement of index of refraction of the water solution, and therefore sensitive to the solute concentrations. How sensitive are the Brix measurements to turbidity or scattering materials in the solution? Is that a possible problem for your measurements of fruit at various stages of maturity? Regards, Dave | ||
| kerry walsh (walshk) Junior Member Username: walshk Post Number: 7 Registered: 2-2007 |
Dear Rui (at the risk of continuing to mix the mammalian culture and fruit composition threads) I concur with Bob and Andrew, but add the following.. - there is dedicated in feld (hand held)instrumentation ans well as in-line (on-pack-line) instrumentation - in some fruits acids are the dominant contributor to Brix, sugars the minor - think limes - if a fruit loses water during storage, its Brix and Dry matter will increase. So in that sense a NIR measure of Brix that is really based on a measure of water should still work. However, something else about teh fruit may have changed - and the optical properties of the skin or tissue may have changed. This will impact your models prediction, particularly in terms of bias. - to accomodate this you must ensure your calibration set includes fruit from all conditions to be tested - bias correction is an accepted strategy, however for QA purposes you should keep an eye on SEP also - the literature is heavily populated by studies in which a researcher got access to an instrument, collected spectra and attribute values from a population, divided the populatio into a cal and val set, ofen deliberately ensuring the val set was very representative of the cal set - such reports fail to justify the wavelength range used, the optical geometry, etc. (general wisdom is that penetration of some cm is needed to get a representative measure of teh fruit, SWNIR (Herschel) is needed, and a partial transmittance (definately not refelectance) geometry - the model should be tested on truly independant populations, as you are doing!! - for a parallel exercise (prediction across many populations, see some of our (Walsh, Subedi) publications in Postharvest Biology and Technology and JNIRS Kerry [email protected] | ||
| Rui Manuel Farinha das Neves Guerra (rmfguerra) New member Username: rmfguerra Post Number: 3 Registered: 6-2011 |
Hi Andrew and Bob, Thank you for you inputs. They were very enlightning. I understand that Brix may be estimated from its correlation with water content, on the condition that "the acids and 'other' remains pretty constant" (quoting Bob). I am trying to get a good model for 'Rocha' pear, a portuguese variety, and it seems to me that there is not a consistent relation between sugars, water and 'others' for: -fruits coming from watered and non-watered orchards -fruits cold stored one month and fruits cold stored three months -fruits stored on controlled atmosphere conditions and fruits stored on normal atmosphere -and so on... For example, if you have two fruits with same Brix and water content, but store them in different conditions, it is very likely that they suffer from different de-hydration effects after some time, changing the correlations amongst Brix components. So, at this point I am thinking if it makes sense to define different calibrations for different conditions (freshly harvested fruits; short term strored fruits; long term stored fruits, etc.) and apply the correct model in each case. The interesting part is that the models usually give acceptable predictions when the values are in the auto-scaled form (in standard deviation units from the mean, which is zero in this case). That is: the models work reasonably in relative terms: starting from an unknown lot of fruit, usually I can predict which are the sweeter fruits and which are the less sweet, independently of the average value of Brix for that lot. However, if I make the predictions in the Brix scale, I may get a significant shift (in the sense of translation) between real and predicted values. At this point a trick may be introduced: recalibration. Simply pick some fruits from the unknown lot and determine destructively its Brix. Assume this is the mean Brix of the unknown lot and correct for the shift described above. However, I don't feel very confortable with this recalibration method - it feels like "cheating". I am wondering also if PLS is really a good technique for very low causality datasets, as Brix may be the case. Using too many variables, most of them with meaningless information, may be a real problem. I feel that in the calibration step, PLS "twists" the loadings of variables with little information in such a way to fit random data patterns. When applied to an independent set, those variables, even with minute loadings, will distort the predictions. Of course, we may choose the better variables...but again, the choice depends whether the fruits are coming from watered or non watered orchards, or from one month or three months storage, blabla... Well, I apologize for this long post. These are my current scientific frustrations! If somebody can add/help/discuss something more I would be very grateful, Thanks again, Rui | ||
| Pierre Dardenne (dardenne) Senior Member Username: dardenne Post Number: 71 Registered: 3-2002 |
Bonjour Didier, I believe you are located at 70 km from us. If you want, you can contact me, we have a little experience in NIRS and I would be glad to exchange about. ([email protected] ). Cordialement, Pierre | ||
| Bob Jordan (jordan) Junior Member Username: jordan Post Number: 7 Registered: 3-2003 |
I think there is an additional factor here with the Brix measure. If the fruit composition is broken down into the basic components for a ripe fruit you see water (easily measured by NIR), sugars (the dominant carbohydrate), the acids and 'other' (both these pretty minor in quantity terms). If the acids and 'other' remains pretty constant then a good measure of Brix can be had directly from the water fraction (Brix=100-water-constant). Indeed Andrew and I joke that NIR in the Herschel range can only measure water. Now look at an unripe kiwifruit or apple and you will find quantities of starch in place of some of the sugar. Starch is structurally very similar to sugar in an NIR sense. So now the NIR might measure well the Dry matter (100%-water), but will fail to measure the Brix as it has difficulty distinguishing sugars from starch (together the dominant carbohydrate). Interestingly the estimate of starch+sugar would work well and this is the basis of NIR being good at predicting ripened Brix - the Brix expected when all the starch has converted to sugar. There is a small hydrolysis component, but that is another story. I think this sort of argument can be used for other situations as well. Bob J. | ||
| Andrew McGlone (mcglone) Senior Member Username: mcglone Post Number: 27 Registered: 2-2001 |
To Rui We (our sensors team here) worked on NIR and brix measurement of fruit for a few years in the late 90's to mid-2000s. With very good success using inexpensive silicon sensors (400 to 1100nm). Our work was commercialised for fast on-line fruit grading in 2001 and the company (Taste Technologies Ltd, NZ) is still in business, still growing, some 10 years later. And other graders companies have similar technologies too. So brix prediction using NIR, at least in the herschel range (750 to 1100 nm), is a well proven techique. However it certainly does not always work out well, and we felt we understood one of the reasons why. Our belief is that you can only accurately predict brix in fruit if the brix is at least moderately correlated with water content. It makes reasonable sense when you get accustomed to the nature and original of spectra in the herschel range. We tried to prove the theory experimentally a couple of times, but largely failed to entangle the matter, and the best we could report was only circumstantial evidence (hence I say it is a belief, not a definitive fact). I presented some of that evidence in a conference paper in 2005 (McGlone et al, Proceedings of NIR 2005, Auckland, NZ). One thing to remember is that brix is a measurement of soluble solids, not just sugars, and acids can contribute. Nevertheless with most fruits the acids are so much lower than the sugars in concentration that their variation has only marginal affects on the brix predictions. | ||
| David Didion (david_umons) New member Username: david_umons Post Number: 3 Registered: 12-2011 |
I am very grateful to all of you!!! Dissolved glucose is certainly not the easiest component to measure with NIRS. Nevertheless my prediction results associated to aqueous solutions are not that bad. Results are far less satisfactory when I build predictive models from spiked culture medium. Obviously it is due to the complexity of the culture medium. It is my problem. Why choosing the glucose (and the lactate)? My objective is to monitor a cell culture in real-time in order to design a supplementation strategy to optimize the protein production. Therefore I need to measure components that are relevant or limiting (nutriments, toxic metabolites, essential amino acids...). The common source of carbon (glucose) and the associated toxic product (lactate), the common source of nitrogen (glutamine) and the associated toxic product (ammonium) have been largely investigated. We think that the monitoring of amino acids is metabolically relevant. I have been told that glucose and lactate are spectroscopically friendly compared with ammonium and amino acids. It is the reason. By lactate I mean the salt that is produced by the cells. Nevertheless For my NIRS experiments I use lactic acid. I think that lactate and lactic acid present the same NIR signal. Am I wrong? Regarding the need for more than two components. I have designed my calibration set looking at breaking the colinearity existing between the glucose and lactate concentrations. I would be delighted to share the file presenting my first PLS1 models. Thank you again! David | ||
| Sheelagh Halsey (shalsey) Advanced Member Username: shalsey Post Number: 24 Registered: 12-2006 |
David There have been a lot of publications on this subject, I have listed some below: Arnold, S.A., Crowley, J., Vaidyanathan, S., Matheson, L., Mohan, P., Hall, J.W., Harvey, L.M., McNeil, B. At-line monitoring of a submerged filamentous bacterial cultivation using NIRS Enzyme and Microbial Technology 2000 27(9) 691- Arnold, S.A., Crowley, J., Woods, N., Harvey, L.M., McNeil, B. In-situ NIRS to monitor key analytes in mammalian cell cultivation Biotechnology & Bioengineering 2003 84(1) 13-19 Blanco, M., Peinado, A.C., Mas, J. Monitoring alcoholic fermentation by joint use of soft and hard modelling methods Analytica Chimica Acta 2006 556(2) 364-373 Card, C., Hunsaker, B., Smith, T., Hirsch, J. Near-Infrared Spectroscopy for Rapid, Simultaneous Monitoring of Multiple Components in Mammalian Cell Culture Bioprocess International 2008 6(3) 58-65 Chung, H., Arnold, M.A., Rhiel, M., Murhammer, D.W. Simultaneous measurements of glucose, glutamine, ammonia, lactate and glutamate in aqueous solutions by NIRS Applied Spectroscopy 1996 50(2) 270-276 Finn, B., Harvey, L.M., McNeil, B. Near-infrared spectroscopic monitoring of biomass, glucose, ethanol and protein content in a high cell density baker's yeast fed-batch bioprocess Yeast 2006 23(7) 507-517 Hagman, A., Sivertsson, P. The use of NIR spectroscopy in monitoring and controlling bioprocesses Process Control & Quality 1998 11(2) 125-128 Li, Y., Brown, C.W., Sun, F-M., McCrady, J.W., Traxler, R.W., Lo, S-C. Non-invasive fermentation analysis using an artificial neural network algorithm for processing NIR spectra Journal of NIR Spectroscopy 1999 7(2) 101-108 Mattes, R.A., Root, D. Prediction of key analytes in mammalian cell culture media with on-line near infrared spectroscopy NIR News 2008 19(2) 9�11 McShane, M.J., Cameron, B.D., Cot�, G.L., Spiegelman, C.H. Improving complex NIR calibrations using a new wavelength selection algorithm Applied Spectroscopy 1999 53(12) 1575-1581 Petersen, N., �dman, P., Padrell, A.E.C., Stocks, S., Lantz, A.E., Gernaey, K.V. In situ near infrared spectroscopy for analyte-specific monitoring of glucose and ammonium in streptomyces coelicolor fermentations Biotechnology Progress 2010 26(1) 263-271 Riley, M.R., Arnold, M.A., Murhammer, D.W., Walls, E. L., DelaCruz, N. Adaptive calibration scheme for quantification of nutrients and by-products in insect cell bioreactors by NIR spectroscopy Biotechnology Progress 1998 14(3) 527-533 Riley, M.R., Crider, H.M. The effect of analyte concentration range on measurement errors obtained by NIRS Talanta 2000 52(3) 473-484 Roychoudhury, P., O�Kennedy, R., McNeil, B., Harvey, L.M. Multiplexing fibre optic near infrared (NIR) spectroscopy as an emerging technology to monitor industrial bioprocesses Analytica Chimica Acta 2007 590(1) 110-117 Tamburini, E., Vaccari, G., Tosi, S., Trilli, A. NIRS: a tool for monitoring submerged fermentation processes using an immersion optical-fibre probe Applied Spectroscopy 2003 57(2) 132-145 Vaidyanathan, S., Harvey, L.M., McNeil, B. Deconvolution of NIR spectral information for monitoring mycelial biomass and other key analytes in a submerged fungal bioprocess Analytica Chimica Acta 2001 428(1) 41-59 Yano, T. Measurement and control of biological processes unsing NIR spectroscopy - measurement of concentration of glucose, glutamine, ammonia, lactic acid and antibody 10th Anniv. Symp., Tsukuba, Japan, 16th to 18th Nov., 1994 1994 131-136 This is certainly an application that NIR is capable of, but it does take a lot of calibration effort as the matrix is complex. Sheelagh | ||
| Karl Norris (knnirs) Senior Member Username: knnirs Post Number: 66 Registered: 8-2009 |
David, I agree with Mark that measuring glucose in a water medium is much more difficult than measuring it in a dry form, but this should not preclude using it as you suggested. I have an interest in glucose detection, and I am glad to try and help you. Can you email me at [email protected], and send the file you discussed? Karl | ||
| Rui Manuel Farinha das Neves Guerra (rmfguerra) New member Username: rmfguerra Post Number: 2 Registered: 6-2011 |
Hi Gabi, Thanks for your answer. I will keep trying to improve my calibrations, then. If you can, please just tell something more: the reliable models you have built/you know from others, are all on the NIR (900 nm-on) or are there any working models in the Vis-NIR (400-1100 nm)? The relevance of the question comes from the cost of the associated spectrometers. Thanks again, Rui | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 73 Registered: 5-2009 |
Rui, It is interesting that you don't find reliable stuff in the literature - because most of the time the literature is more optimistic than reality - particularly when reporting measurements of food products. To brix itself - it is measurable and the degree of reliability is not because sugars (glucose, fructose and sucrose have exactly same spectrum when in water) don't have features that allow their measurement. the reference brix reflects more than just the sugars - and the "other" contributions to reference values may vary substantially between one product to another and these variations are reflected in the spectrum and "cause confusion" - but in most cases the calibrations can overcome many difficulties - rendering the practical measurement of brix by NIR very useful in many industries, inclusing fruit analysis. I personally have many well working applications in fruits, juices etc. I hope it helps a little bit Gabi Levin Brimrose | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 72 Registered: 5-2009 |
Hi David, By lactate - you mean lactic acid? lactic acid salt? which salt? I agree fully with Howard - glucose is the least ineteresting to get training with - it has a minor effect in the solution in water on the spectrum. It is measureable in aqueous solutions but less friendy for training. Lactic acid salts aren't great either for this purpose, although they contain C-H bonds. Since you need to operate in aqueous solutions in the final run - it is better to use ethanol, methanol, propanol and possibly add methyl amine which is also soluble in water. The use of more than two components is very important - as it allows you to create concentration variations that avoid linear interdependency between concentrations of the species. This will also help you identify the locations of peaks of C-H bonds, peaks of O-H bonds in compounds other than water, peaks of N-H bonds that could be - but not fully related to peaks in the amino acids. I hope this contributes to your work. Gabi levin Brimrose corp. | ||
| Rui Manuel Farinha das Neves Guerra (rmfguerra) New member Username: rmfguerra Post Number: 1 Registered: 6-2011 |
Dear all, This is also my first post, although I have been reading the discussions so some time now. This discussion makes me remember that there are lots of studies about measuring the Brix of fruits with Vis/NIR and/or NIR spectroscopy. The Brix degrees are basically proportional to the content of sugars dissolved in the fruit juice. The idea is to measure it non destructively through reflectance or transmittance spectroscopy. But, quoting Howard Mark "Glucose is not a particularly NIR-friendly compound" - and probably not the other forms of sugar present in fruits. I have been doing work on that, but the models do not seem very reliable. Even with supposedly large and representative calibration sets, it is common to test an external validation set and get very bad predictions. On the other hand, what I see in the literature is never reassuring: small calibration sets, the models are not tested against truly independent validation sets, etc. This seems to point to the fact that most of the time I am modelling indirect correlations with sugars and not sugars themselves. So, and once again, having in mind what Howard Mark said about glucose, what do you think about the possibility of building a reliable NIR model for Brix in fruit? Thanks, Rui | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 482 Registered: 9-2001 |
Come to think of it, you may also want to check the structure of Lactate before you do any extensive amount of work with it. \o/ /_\ | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 481 Registered: 9-2001 |
David - welcome to the discussion group. Unfortunately, your choice of trial compounds was a little off the mark. Glucose is not a particularly NIR-friendly compound. In the solid state it shows some nice peaks, but when it's dissolved, all those peaks disappear due to their association with water. For an NIR-friendly material, you need to use something with a good amount of -CH3 in it, and glucose does not have that. Howard \o/ /_\ | ||
| David Didion (david_umons) New member Username: david_umons Post Number: 2 Registered: 12-2011 |
Dear all, I am a PhD student from Belgium looking at monitoring mammalian cell cultures with NIRS. Target components are amino acids but since my department has no background in NIRS application, I start with "spectroscopically friendly" components like glucose and lactate. I don't succeed in building satisfactory predictive models for the quantification of glucose and lactate in the cell culture medium. By "satisfactory" I mean approaching the accuracy and the precision that I find in the literature. Particularly it is the lack of precision (repeatability for triplicate measurements) and the general dispersion on both sides of the bisector that surprise me. I have tried to attache a file full of details to facilitate the discussion but it far exceeds the size limit. I will be pleased to send you this file to discuss my application and my analysis parameters. I use a Bruker Matrix-F FT-NIR and a Hellma Immersion Transflectance probe. Thank you in advance! David |