search

CLINDA-AAU / hemaClass-paper

Paper for the hemaClass package/website
3 stars 1 forks source link

The CHO classifier #16

Closed falgreen closed 9 years ago

falgreen commented 9 years ago

The CHO classifier is already part of the REGS article. I think we should make this clear and use same notation as in the original article, i.e. capital letters.

Perhaps the derivation of Graham's formula could be kept on the homepage instead of being a supplementary to the article.

AEBilgrau commented 9 years ago

Agree on the reference (now added). However, there are some reasons the notation changed:

  1. The REGS paper uses the probability of resistance, whereas the website and package use the probability of sensitivity. So we avoid two different, inverse meanings to the same symbol across papers.
  2. If you denote by C, H, and O the probability of sensitivity toward the drugs, then CHO can be misinterpreted as the product of these probabilities. So, using CHO as the probability of sensitivity toward the combination is ambiguous. This is solved by the indexing.
  3. Strictly, the probability of being sensitive towards the combination, is not the same as simulatenous sensitivity the three, i.e. P(C, H, O). In other words: P(sensitive to C and H and O) != P(sensitive CHO combination). There is a subtle difference as interaction effects of the drugs may be present.
falgreen commented 9 years ago

I think there is some misunderstanding here. In the reference we already use indexing. This is the section from the article:

By use of Graham’s formula the HBBCL based REGS classifiers for C, H, and O were combined into a single REGS classifier for CHO. Let $P_C$, $P_H$, and $P_O$ denote the probabilities of being resistant towards the three drugs C, H, and O individually. Then under the assumption of drug independence the posterior probability of being resistant towards the combination therapy CHO was estimated as: $P_C P_H P_O/(P_C P_H P_O+(1-P_C)(1-P_H)(1-P_O))$

falgreen commented 9 years ago

By capital letters I meant capital P's

falgreen commented 9 years ago

Btw is there a good reason for why we have opted for sensitivity rather than resistance (It would be easy enough to change the homepage).

mboegsted commented 9 years ago

I have no good reasons for that. Steffen, you are actually the one who changed from resistance to sensitivity, if you have no good reasons, I really think we should change it all.

mboegsted commented 9 years ago

It would really be nice as it then complies with the REGS and BAGS papers

AEBilgrau commented 9 years ago

Arh, sorry. I just thought about the first draft of the Graham derivation (which I thought followed the notation in REGS) which used the drug letters for the probabilities. Anyway, capital P's or lowercase p's makes no difference to me.

falgreen commented 9 years ago

Then let us use capital. It doesn't make any difference to me either, but I like that it is the same in the two articles

Den 23. okt. 2015 kl. 09.32 skrev Anders Ellern Bilgrau notifications@github.com:

Arh, sorry. I just thought about the first draft of the Graham derivation (which I thought followed the notation in REGS) which uses the drug letters for the probabilities. Anyway, capital P's or lowercase p's makes no difference to me.

— Reply to this email directly or view it on GitHub.