The use of self-organizing maps in the study of phytotoxic activity of Salvia species (Lamiaceae).

A screening study on the anti-germination activity of the fresh aerial part plant exudates of thirteen species of Salvia L.(Lamiaceae) against Papaver rhoeas L. and Avena sativa L. in Petri dish and pot experiments, evaluated on the basis of statistical indices both classical ones [1, 2] and proposed by our group has been carried out. All indices has shown that the greatest part of the Salvia exudates possess inhibitory activity on the two species germination and may be considered as promising mixtures of phytotoxins with a potential for development of natural product herbicides. Nevertheless, clear differences among the majority of the Salvia species are not highlighted. For this reason, a clustering algorithm based on Self-Organizing-Maps (SOMs) has been used to verify if this approach can single out small differences [3] that are not evident by classical methods. Mathematical elaborations have been applied before clustering to correctly feed the network. Raw data are passed to SOMs as matrix, so that if columns/rows contained too similar data, these columns/rows gave too few information to the process. A selection to these elements has been applied fixing some constraints about maximum and minimum of contained values, Variation Coefficient [4], and correlation between data columns. After elaboration, we clustered data for both target species and in both conditions getting four sets of results. In all these sets, the macroscopic evidences singled out by statistical indexes are present, but small difference can be qualitatively defined up to find 5 clusters in the Salvia species.