Autogating em dados de citometria de fluxo utilizando classificadores SVM para identificação de bacterioplâncton

Abstract

This master tesis shows the proposal to develop a methodology - together with the presentation of the results of its application - that uses a machine learning technique, SVM, for automated analysis of flow cytometry data in samples of aquatic environments, identification of bacterioplankton. The samples used in the execution of this methodology were collected in 19 high altitude mountain lakes that were manually classified in the Laboratory of Limnology of the Department of Oceanography and Limnology of UFRN. Previously, it started with some tests of kernel configuration and a quantitative analysis based on the average number of hits in the automated classification, in which it was noticed that the prediction error rate varied between 1.86 % and 3, 35 % on average. Two stages of development of the proposed methodology were carried out, where prediction models were created and a series of tests were carried out with the databases created from the available information. The results were exposed to a series of quantitative and qualitative analyzes, including using PCA to understand the importance of each variable in the sample data sets. For a qualitative evaluation of the proposed methodology, a statistical analysis was applied to compare both strategies of prediction models, which is based on the final classification indicated by the algorithm of Support Vector Machine.