Slow down, you move too fast […] Feelin’ Groovy [Bioanalytics]

posted by Helmut Homepage – Vienna, Austria, 2018-11-06 03:05 (2228 d 08:08 ago) – Posting: # 19537
Views: 10,587

Hi ElMaestro,

❝ I am inclined to do this: [some R-code]


❝

❝ where the red line indicates the level of noise (in this case, right of the peak) and the greeen one is the signal. Roughly.


Disagree and agree.

❝ Note that in both cases I quantify s as well as n in one direction from the baseline mean or median or whatever.

❝

❝ Thus I am landing at s:n = (18-4.1) / (8-4.1) = 3.6.

❝ I am not in any way claiming this is better or worse, only that this is my idea of an approach.


OK, you are aware that this is not according to the textbooks about chromatography.1 See also Dolan’s figure 1.2 Noise is defined as the range of recorded “zero” signals. Hence, your S:N is by a factor of two better than mine.

❝ If I recall correctly, if you are "a large software vendor" -and I will mention none in particular- you can also do something like:

❝

❝ k=3                   #a miserable sad pointless constant to make s:n look better ??

❝ a=sd (y3[400:714])    #sd of points on the peak

❝ b=sd (y3[800:1000])   #sd of points adjacent to the peak

❝ sn=k*a/b

❝

❝ which gives a result of about 5.:-D


Well…

❝ Personally, I would of course always adjust k so that s:n is not less than 10 or so, just to avoid questions. I mean, I care about my data because I am not a nasty person :-D:-D:-D


Well, your example is artificial. Remember this one? What you constructed here is what you would get when you sample from the A/D-converter at 1.66 Hz. For your peak width that’s by a factor of five to ten too fast. With a reasonable bundling you would get the blue line (no time-constant; moving averages for simplicity) …

[image]


… and, voilà, an S:N of ~4.7. Signal bundling is not done for fun. Peak detection (start/end) requires the 1st derivative of the signal and finding the “peak’s summit” the 2nd. See these examples (slides 11–12). Difficult enough for bundled data, impossible for what you consider “raw data”. It’s simply a necessity for proper integration. Finding the right value is always a compromise. If you bundle too much, peaks will be distorted and separation between adjacent peaks suffers. If you bundle too little, the system will have a hard time detecting the right start/end of peaks. Nothing is perfect.

  1. Kuss H-J, Kromidas S (eds.): Quantification in LC and GC. Weinheim: Wiley-VCH; 2009.
  2. Dolan JW. The Role of the Signal-to-Noise Ratio in Precision and Accuracy. LCGC North America. 2005:23(12);1256–1260.
—
Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

UA Flag
Activity
 Admin contact
23,336 posts in 4,902 threads, 1,666 registered users;
40 visitors (0 registered, 40 guests [including 13 identified bots]).
Forum time: 11:14 CET (Europe/Vienna)

Biostatistician. One who has neither the intellect for mathematics
nor the commitment for medicine but likes to dabble in both.    Stephen Senn

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5