Optimal segments: a method for smoothing tracer data to calculate metabolic fluxes

Abstract

Metabolic fluxes (Ra and Rd) are calculated in the nonsteady state using Steele's equations. These calculations require estimates of the values and rates of change of glucose and specific activity at discrete sampling times. Data smoothing minimizes the effect of measurement error that confounds the Ra and Rd calculations. We compared three smoothing methods: a) moving average, b) polynomial fitting, and c) optimal segments, a new technique that utilizes optimization methods. Experimental designs were simulated: 1) constant infusion of glucagon in depancreatized dogs, and 2) glucose oscillations generated by constant high-level glucose infusion. Measurement error was added to raw data. After smoothing, fluxes were calculated and compared to the "actual" Ra and Rd. Ra calculated from unsmoothed noisy data were in error by an average 39%. Error was reduced by smoothing to: 23%, moving average; 18%, polynomial fitting; 15%, optimal segments. Optimal segments was best for calculating Ra (P less than 0.01) and was better than or equal to other methods for Rd. Distortion in flux patterns was greatest for polynomial fitting (P less than 0.01) and least for optimal segments (P less than 0.001). Optimal segments is the method of choice for smoothing tracer data; it improves calculations of Ra and Rd with minimal distortion.