Please kindly answer my question again (screening)
To Dr. Phil Williams
From: panmanas sirisomboon (panmanas@hotmail.com)
Sent: Wednesday, March 03, 2010 4:51:22 PM
To: Dr. Phil Williams (philwilliams@pdkgrain.com)
Dear Dr. Phil
How are you? Long time I didnt talk to you. I has no question along my way but today I and my students have one question.
I would like to ask you that in Page 5-8 of your book, in Table 5.2.5.1, what do you mean by the word "screening". Could you please kindly give us an example.
Could you please kindly clarify for me and my students.
Thank you very much indeed.
Respectfully yours,
Panmanas
Re: Please kindly answer my question again (screening)
To panmanas sirisomboon
From: Phil Williams (philwilliams@pdkgrain.com)
Sent: Wednesday, March 03, 2010 7:04:38 PM
To: panmanas sirisomboon (panmanas@hotmail.com)
Hello Panmanas, good to hear from you, as ever. I am well, thankfully, and very busy editing the Proceedings of NIR-2009. I have now reviewed, edited, and in a lot of cases partially re-written over 130 papers, with about another 70 to do. I am also learning things about new applications, so it is an interesting time.
Screening means identifying the best (and worst) individuals from any type of population. It is a process widely-used in plant-breeding. Plant breeders, whether they are working with wheat, barley, rice, chickpeas, potatoes, or any crop that is to be grown for food or feed need to be able to identify individual plants that are of the best, and the worst quality for processing, or direct eating (we use "food" for things that are intended to be eaten by people, and "feed" for things that are mainly intended to be eaten by animals). Most crops, such as wheat or rice, are not eaten directly, but have to be processed in some way before being prepared into foods, so processing quality is important. Texture (degree of hardness or softness in grains, "firmness" in fruits and root crops, and "toughness" in fibrous crops) has a very strong influence on processing quality. The process of identification is called "screening".
Plant breeders produce 1000s of new lines every year in early generations. Testing for constituents such as protein, moisture, oil/fat, starch and sugar contents by reference methods is expensive and time-consuming, and as a result most of the early generation material is advanced on the basis of resistance to disease, which can be quickly evaluated (screened) in the field, and yield, rather than on the basis of quality. Testing for quality is postponed until later generations, when there are far fewer lines to test (and less expense involved). As a result a lot of advanced new breeders' lines are found to be deficient in quality after a lot of time and expense has been invested over several seasons. This is of course a waste of both money and time, and is frustrating to the breeder (as well as disappointing to them, because they are people with feelings, and dreams of success).
NIRS has provided a simple and cheap (inexpensive) method for testing all of the lines produced by plant breeders, and even the early generation lines can be tested for the most important constituents and parameters. Testing for processing quality, such as texture, and organoleptic parameters, such as taste is not so straightforward (easy) as testing for constituents, because there are no absorbers that make this type of testing directly measurable by NIRS. This is one of the reasons why I introduced the two tables of guidelines for the RPD. It should be possible to obtain RPD values of 4 and higher for NIRS testing for constituents, because the absorbers (C-H, O-H, N-H, etc.) that are directly associated with the constituents are present. It is more difficult to get such high values for NIRS testing for the more difficult things, such as texture and digestibility, so I dropped the RPD value for rough screening to 2.0. What an RPD of 2.0 means is that the NIRS SEP results can be used to sort out the best and worst of the population, which is at least better than not screening for quality at all. Obviously, the higher the RPD, the lower will be the SEP relative to the overall variance (SD) of the entire population, and the screening will be the more reliable. If the RPD is much lower than 2.0 the efficiency of the screening will be reduced, and lines that are identified as being of higher quality should be re-tested and verified, where practicable.
I hope that this helps, Panmanas, but please don't hesitate to ask me further questions if I haven't answered you fully here. I wish that all of the University teachers who have attended one of my courses used the course as well as you do, it is a pleasure to work with you.
Sincerely,
Phil.
RE: Please kindly answer my question again (screening)
To Dr. Phil Williams
From: panmanas sirisomboon (panmanas@hotmail.com)
Sent: Thursday, March 04, 2010 1:28:31 AM
To: Dr. Phil Williams (philwilliams@pdkgrain.com)
Dear Dr. Phil
I am very glad to recieve your reply. Thank you very much indeed.
You enjoy working. You are quite busy and you make me would like to read the proceedings very much.
It must be very good book.
I study what you teach and I understand more about screening.
Your words I can use as a reference words when I report my result in a paper. But still I doubt about how can we conclude our model is for screening because we are doing quantitative by predicting the constituents and then we get RPD and r value and from that we can compare the result with your guidance in tables and then we conclude the our model is for screening.
Please again kindly reply for me and my students.
Thank you very much indeed.
Respectfully yours,
Panmanas
Re: Please kindly answer my question again (screening)
To panmanas sirisomboon
From: Phil Williams (philwilliams@pdkgrain.com)
Sent: Thursday, March 04, 2010 8:43:21 PM
To: panmanas sirisomboon (panmanas@hotmail.com)
Hello Panmanas, the RPD guidance table in the NIRS book seems to have confused you a bit, and I am sorry about that. Something that I said to a scientist from New Zealand that came to work with me about 30 years ago made him very happy, and I have used it many times since. It is "If it works for you, use it"! If you are happy with your results and are confident that your model is predicting what you want it to predict effectively, use it. Don't worry about "screening". Screening essentially means approximate testing (estimation), rather than very precise testing. You only need the accuracy/precision that you need to get the information necessary to get the job done. For example, there is no need to try to get accuracy of 0.1% if 0.2 or even 0.3% is good enough.
From what I have learned in editing nearly 150 papers from NIR-2009 I feel that it would be good if someone like me could come to work with you and Khun Som at your universities for a few months, also some universities in China, and even in Japan.
Keep smiling,
Phil.
----- Or
RE: Please kindly answer my question again (screening)
To Dr. Phil Williams
From: panmanas sirisomboon (panmanas@hotmail.com)
Sent: Friday, March 05, 2010 2:11:24 AM
To: Dr. Phil Williams (philwilliams@pdkgrain.com)
Dear Dr. Phil
I am very glad that you are interested in working with us in Thailand. Thank you very much indeed.
I would like you to come and advise me and do research with me in my laboratory. I shall try to find scholarship for you to come. In my KMITL there is guest rooms in condominiam that you and your family can stay. You will enjoy Thai weather. And we can learn a lot from you and improve our works on NIRS. Please make my dream come true.
Thank you very much for your teaching again. I am happy to use your guidance tables or values. We can determine the result whether it is good or it have to be improve or not. what I dont understand is why quantitative can be qualitative.
So now I will consider the word "screening" to be "roughly estimation"
Please have merry time.
Respectfully yours,
Panmanas
Re: Please kindly answer my question again (screening)
To panmanas sirisomboon
From: Phil Williams (philwilliams@pdkgrain.com)
Sent: Friday, March 05, 2010 5:40:27 PM
To: panmanas sirisomboon (panmanas@hotmail.com)
1 attachment (21.0 KB)
Extended ...doc
View online
Download(21.0 KB)
Table 5.20 The RPD and RER Statistics. Grains, flours, constituents
RPD Value RER Value Classification/usefulness Application
0.0 - 2.3 Up to 6 Very poor Not recommended
2.4 - 3.0 7-12 Poor Very rough screening
3.1 - 4.9 13-20 Fair Screening
5.0 - 6.4 21-30 Good Quality control
6.5 - 8.0 31-40 Very good Process control
8.1+ 41+ Excellent Any application
Near-infrared spectroscopy has been successfully applied to a very variable range of commodities, constituents and functionality factors. Because of complications with sample preparation, sample presentation to the instrument and difficulties with reference testing RPD values of as high as even 3.0 may be difficult to achieve. But even with lower values for the RPD or RER, these statistics still give a valuable indication of the evaluation of applications of NIRS for these types of industrial and scientific analytical work. Table 5.21 gives the interpretation of RPD and RER values for some different and more “difficult” applications, such as analysis of forages, high moisture materials, such as vegetables, fruits, meat and fish, soils and manures, where the application of NIRS is affected by the more complex nature of the materials.
Table 5.21 The RPD and RER Statistics. Forages, feeds, functionality factors
RPD Value RER Value Classification/usefulness Application
0.0 – 1.9 Up to 5 Very poor Not recommended
2.0 – 2.4 6-8 Poor Rough screening
2.5 - 2.9 9-11 Fair Screening
3.0 - 3.4 12-14 Good Quality control
3.5 - 4.0 15-18 Very good Process control
4.1+ 18-20 Excellent Any application
Table 5.21 is also applicable to the prediction of functionality factors such as grain texture, digestibility, and energy. The different interpretation does not change the statistical consequences of the lower values, but it places them in a category of explanation that is more realistic in terms of the results likely to be achieved with these applications.
In terms of qualitative and quantitative analysis by NIRS they can be compared this way: with high values of RPD quantitative analytical results can be expected that would be good enough for quality control in a factory, and qualitative results in terms of classification that would identify classes correctly 90 % of the time. With RPD values in the "screening" area (e.g. a RPD of 2.5-2.9 from Table 5.21) quantitative analysis results can be expected that would estimate composition or functionality well enough to identify high, medium and low levels, and qualitative results that would identify classes correctly about 66-70 % of the time, probably still better than methods used before NIRS, and better than doing no testing at all.
Panmanas, I am adding the page that contains the RPD guidance table, which is now expanded to Tables 5.20 and 5.21. I added Table 5.21 to encourage people with difficult applications that even RPD values as low as 2.0 can still be of some use, and often as good as the methods that would be used if there was no NIRS. In terms of qualitative and quantitative analysis by NIRS you can compare them this way: with high values of RPD you can expect quantitative analytical results that would be good enough for quality control in a factory, and qualitative results in terms of classification that would identify classes correctly 90 % of the time. With RPD values in the "screening" area (e.g. a RPD of 2.5-2.9 from Table 5.21) you could expect quantitative analysis results that would estimate composition or functionality well enough to identify high, medium and low levels, and qualitative results that would identify classes correctly about 66-70 % of the time,again probably still better than methods used before NIRS, and better than doing no testing at all.
Your questions are useful to me. When I answer them they sometimes make me think of an area where the course can be improved, so I add the new points. This is how Table 5.21 was developed, and I added the last paragraph just now, after I had written this message to you, so keep the questions coming! I am in the process of up-dating the course (after I have finished editing the papers for NIR-2009 Proceedings). When it is done I will send you the up-dated version, in the same way as I sent you Section 10 last May 17th. Even Section 10 will be up-dated this year.
Greetings,
Phil.