Supervised Identification of Acinetobacter Baumanni Strains Using Artificial Neural Network

Abstract

In hospital environments around the world bacterial contamination is prevalence. One of the most commonly found bacteria is the Acinetobacter Baumannii. It can cause unitary tract, lung, abdominal and central nervous system infection. This bacteria is becoming more resistant to antibiotics. Thus, identification of the non-resistant from the resistant bacteria strain is of important for the correct course of treatments. We propose to use the artificial neural network (ANN) for supervised identification of this bacteria. The mass spectra generated from the liquid chromatography mass spectrometry (LCMS) were used as the features to train the ANN. However, due to the massive number of features, we applied the principle component analysis (PCA) to reduce the dimensions. Less than 1% of the original number of features were utilized. The hand out validation method confirmed that the accuracy, sensitivity and specificity are 0.75 respectively. In order to avoid selection biasness in the sampling, 5-fold cross validation was performed. In comparison, the average accuracy is close to 0.75 but the average sensitivity is slightly higher by 0.50