Optimal bandwidth selection in stochastic regression of Bio-FET measurements

Abstract

Biological field effect transistors (Bio-FETs) are modern bioelectronics instruments that offer rapid, low-cost and accurate point of care (POC) biomarker measurements. The time series data produced by these devices contain noise which interferes with quantitative analysis. Stochastic regression, which relies on modeling the measurement with a linear stochastic drift-diffusion equation with unknown coefficients, is employed to separate signal from noise. Coefficients are estimated through local weighted regression and maximum likelihood estimation, both of which depend on a kernel function and the size of a bandwidth parameter. In this work we determine the optimal bandwidth parameter associated with an experimental Bio-FET measurement, by considering three distinct but related kernel functions. Cross validation is performed with respect to different instrument aspect ratios. Results show optimal bandwidth parameters are surprisingly consistent across aspect ratios, and suggest a choice of kernel function.