Statistical Analysis of results of HPLC and HPTLC

[pjain]

I am working with bioactive and pahrmaceutically active compounds present in medicinal and aromatic plants grown under differential stress conditions. I want to statistically analyzie my results

Which is the best software for statistically analyzing results generated from HPLC and HPTLC ?

What is CRD and RDB statistical design ?

What is SEM, CD in statistics?

Thanks

[tom jupille]

The term "statistically analyze" is so vague as to be useless. What are you trying to accomplish?

Which is the best software for statistically analyzing results generated from HPLC and HPTLC ?

Per my comment above, that depends what you are trying to do. Microsoft Excel includes a wide range of statistical functions.

What is CRD and RDB statistical design ?

CRD = complete randomized design
I'm not familiar with "RDB" as applied to statistical design. Perhaps it refers to a "relational data base"?

What is SEM, CD in statistics?

SEM = standard error of the mean.
I've never seen "CD" used in a statistical context. I can speculate that it refers to the coefficient of determination (correctly shown as r^2).

[freemab]

It sounds like perhaps you are unfamiliar with statistics. If this be the case, I strongly suggest you read up on statistics for analytical chemists, as a knowledge of what you are dealing with could save you much grief and possible false conclusions.

The primary thing to remember about statistics is that it is a tool and only a tool. With rare exceptions, you cannot turn over your data to a software package and expect to get a meaningful interpretation from it. It has been demonstrated that very different data sets can give rise to identical statistics - mean, standard deviation, etc. - giving the impression that there is no difference between them, when nothing could be further from the truth.

In this regard, it is always a good idea to come up with some way of plotting the data, as a graph - literally - gives a picture that a statistical summary never will. For example, suppose you want to compare the amounts of compound A found in 3 plants grown under standard conditions versus 3 plants grown under stressed conditions (whatever those may be). Without regard to any other statistical tests you may do, it is a very good idea to plot these values on two parallel axes. These "dot plots" can give you a good first take on whether there is a difference between the two groups, and whether the difference is meaningful. (3 is not a large number, however. 20 would be much better.)

Similarly, if you have data for the amount of compound A versus increased "stress", and wish run a regression to determine correlation, it is always a good idea to plot the data first. Not only will this demonstrate whether there is any apparent correlation, it may help distinguish what sort of correlation (linear, nonlinear, etc.) there may be.

Finally, the concept of "statistical significance" is paramount. On can see a difference in two average values but determine from tests on the standard deviations that there is no statistically significant difference between these values. Such conclusions can be a real headache. Good judgment is needed. I hard sciences like chemistry, we may test to a "95% confidence", meaning, there's only one chance in 20 that our conclusion is wrong. However, your stress tests of plants are biological, and for a variety of reasons, a 95% test may be far to severe.

[Don_Hilton]

Is the item perhaps CV rather than CD? CV (Coefficient of Variation) is fairly commonly used.

And, rather than try to explain it, I will strongly suggest that Pjain take a class in statistics – or at least obtain a good introductory text.

Knowing how to compute these terms is important, but it is important to know what the null hypothesis is and to have an understanding of what statistics can and can not tell you is perhaps even more critical. Statistical results are no better than the experimental design.

[JGK]

And, rather than try to explain it, I will strongly suggest that Pjain take a class in statistics – or at least obtain a good introductory text.

As an introductory text I can highly recommend

"Statistics and Chemometrics for Analytical Chemistry"
by James N. Miller (Author), Jane C. Miller (Author)

Prof J. N. Miller was the Statistics and Chemometrics Lecturer at the university where I studied for my MSc degree (I passed )

[aceto_81]

I am working with bioactive and pahrmaceutically active compounds present in medicinal and aromatic plants grown under differential stress conditions. I want to statistically analyzie my results

Which is the best software for statistically analyzing results generated from HPLC and HPTLC ?

What is CRD and RDB statistical design ?

What is SEM, CD in statistics?

Thanks

Maybe you should post what is your final goal to met, so that we can give a tailored advise. Best software is hard to say, is GLP certification required or not, is it mostely to generate random designs, or is it to perform ANOVA, do some lineare regression, or maximum likelihood, ....

All depends on your application..

Ace

[yangz00g]

It seems that you are talking about Metabolomics. There are a few commercial available software for this purpose (google it), but IMO, none of them can replace an experienced expert.

[weeker2001]

i think you should read some text about statistics .what you want from you experiment? some software will do you a favor ,but first you need do some preparetion about statistics . SPSS SAS and other software are mostly used in analysis data .

[Don_Hilton]

With the mention of SPSS and SAS, both of which I love, I have to mention one that I've run across reciently and it seems to have great price and functionaly: "R" http://www.r-project.org/

It seems to be accepted widely in the scientific community and costs only the time it takes to download it. However, it does not seem to have the same kind of data preparation facilities as SAS, so you have to have your data ready to go as you get it into R. (I cann't compare SPSS any more. My last use of that was on a mainframe, back in the 1970's.)

And the big thing with SAS, SPSS, R, or any of the other packages: If you make incorrect assumptions about your data set (or fail to know what is being assumed) - you will still get regressions, clustering, and all of that. But your results will be no better than your understanding of what you did.

[aceto_81]

It seems to be accepted widely in the scientific community and costs only the time it takes to download it. However, it does not seem to have the same kind of data preparation facilities as SAS, so you have to have your data ready to go as you get it into R. (I cann't compare SPSS any more. My last use of that was on a mainframe, back in the 1970's.)

+1 for R, it is widely accepted in all areas of scientific computing.
And if you have specific needs, there are packages available for free in lots of areas.
Also you have their suport list, which enables troubleshooting much faster than any commercial package.

But I don't agree for 100% with the statement "you have to have your data ready to go": There are a lot of ways to import your data, in lots of formats. But I must agree: the easiest way is to put your data already in a table.

One minor point: R has a steep learning curve, but once you manage it, there are plenty of option. Also there are GUI's available for R (also for free)

Best regards

Ace

[Peter Apps]

Hi Ace

You get a steep learning curve when a thing is easy to learn - the curve is a plot of success against number of tries. If performance is not improving the curve is flat.

Cheers

Peter

[Don_Hilton]

Ace,

You are correct that there are a number of ways that you can import data into R. I was speaking relative to SAS, which allows you to have the language programmatically open instrument reports -- headers, data columns, and all and all; parse results out of the reports; and build the table for you and check the entries. With checking, it can turn a result entered into a table as "ND" to 0, a missing value, or however you want to tread a non-detect. So far, I've not found the way to deal with that text entry in a numeric column with R.

As systems become more integrated, some of the functionality I've used with SAS on laboratory data will be less often required.

I agree that the learning curve is steep. I've been trying to get started and for effort applied as a function of progress - I've been putting in the effort. (I just need the time to put in more.) And, it is not entirely without progress. The plots are fantastic.

Don

[lmh]

If you want metabolomics and are using R, then of course you should be looking at the xcms package. Fortunately it is easier to google for "xcms" than "r".

[aceto_81]

Also the bioconductor packages may be promising.
And: searching for R: www.rseek.org

Ace