Signal processing for biologically inspired sensors

Abstract

Pores or channels with diameters in the range of nanometers up to micrometers can be used as Coulter counting apertures to detect particles and organic molecules such as proteins. Coulter counting is performed by applying a constant potential across a nano- or micropore while recording the drop in ionic current upon passage of a molecule. Looking at the shape and duration of these current pulses enables us to estimate the size as well as the concentration of these molecules. Discrimination between different analytes can be performed by extracting appropriate features from the Coulter signals (events) and using them for classification. The challenge in being able to identify particular analytes is that a drop in current can also be caused by a molecule bouncing off the pore wall rather than moving through the micropore. Such drops are called non-events and can be discriminated from the events using Support Vector Machines. In this paper, we consider the amplitude of the current drop and the duration of the current pulse as features to determine if an event occurred. The proposed approach uses the Dirichlet process mixture model to cluster the data in the feature domain as the type of the events in the signal record is unknown. Results obtained show that the Dirichlet process mixture model accurately finds the types of events and their count for each signal record.