Differentiation of MRSA using Photoacoustics

In order to interpret the photoacoustic data provided by the flow cytometer, we used k-means clustering to guide our differentiation between methicillin resistant and susceptible samples (28 ). Although a simple interpretation of the number of bacteria after oxacillin treatment compared to the number in the untreated subsample should obviously indicate whether the bacteria was methicillin resistant or not, we used a formal method to automatically determine resistance. K-means, like other clustering methods, takes data points in a space of one or more dimensions and determines natural groupings of those points by proximity. Given a number of clusters, k, the algorithm separates all points into that number of groups. Since we are only interested in resistant and susceptible groups, we chose k = 2. We took the ratio of treated detection numbers over untreated numbers resulting in 13 numbers, approximately in the range of 0 to 1. Lower numbers would indicate that oxacillin was effective in decreasing the S. aureus population. However, there was no ad hoc threshold for determining antibiotic resistance. We applied k-means with two clusters to this data set. We used the Matlab function, kmeans, which uses Lloyd's algorithm. For simplicity, we used Euclidean distance for measuring and establishing iterative clusters. This analysis resulted in two clearly defined clusters for MRSA and methicillin susceptible samples.