B Van Oystaeyen | PubMed Health

PubMed1995

[Helpful to bacterial identification: an alternative to probability calculation].

In the field of bacterial identification, the usual probabilistic approach may be criticized to some extent for the way it deals with weak tests. To improve this, we propose a modification which consists to compare the probability of the observed profile to the average of all the possible answers of a given bacteria. The result is discussed and showed to be understood in terms of plausibility rather than probability.

Bacteria↗

PubMedDate not supplied

[Numerical help to bacterial identification: definition of a first kind error risk].

The numerical method commonly used as an help to the bacterian identification appears quite efficient in its first role, i.e. to propose the name of the plausible bacteria, but it fails to define a confidence level to be assiociated to the final decision. To correct this, this paper propose to calculate the response probability distribution for each possible bacteria. The calculation uses the so-called "Monte-Carlo" method and has to be performed once for all on a given set of biochemical tests. For each calculated ditribution, it is easy to find a first kind error risk, or alpha-error risk. The proposed theoretical approach moreover avoid the comparison of "identification scores", which is always questionable so far the various bacteria present very different sensitivities to the biochemical tests.

Bacteria↗