ABSTRACT
Experimental microbiology yields a huge quantity of raw data which needs to be evaluated and classified in a wide variety of situation from marine research, environmental pollution and pharmacokinetics of antimicrobial agents to epidemiological clinical trials on infectious diseases. It is indispensable in all kinds of disciplines to validate, transform and correlate data clusters to demonstrate the statistical significance of results. Whether academic or biotechnological, the scientific credibility of a work is strongly affected by the statistical methods and their adequacy. For a simple univariate analysis, many commercial or open source software products are available to perform sophisticated statistics for discriminant and multi-factorial analysis, but the majority of scientists use statistics partially. This is due to the high competence level required by a multivariate approach. It is known that the choice of a test, correct distribution assumption, valid experimental design and preliminary raw data validation are prejudicial to good science. All kinds of experimentation need analytical interpretation of descriptive evidence so that a classical numerical approach is not enough when raw data are not validated or incomplete. Microbiologists always wish to quickly discriminate, or correlate, group and data clusters concerning clinical patient profiles, auditing multi-sensor derived numbers, monitoring biosphere indicators on either chemical and physical parameters or dynamics of microbe populations. Mathematical and statistical analysis is essential to distinguish phenotypes or constraints. Data are in general stored in spreadsheet and database files which change continuously depending on the data collection and scope. We here propose a Records Matching Method (RMM) suitable for any kind of cluster analysis and pattern identification which can be used for either parametric or non parametric analysis without necessarily stating the pre-process statistical assumption on variable distribution. The RMM is an application of a theoretical approach based on the Unique Factorisation Domain and is explained with an ideal generalisation model and then applied to a real-world microbiological study. We used an easy mathematical formalism and discuss the possible application of the method as widely applicable to a plethora of taxonomic and phenetic investigations as well as for clinical trials and epidemiology. Prototyping of the model for a computational automated process are also described in order to devise simple software which can infer on data using a heuristic rules file.
