Evaluation of the relationship between peak characteristics and detection performance in two-dimensional...

Peak detection is a critical step of the data analysis workflow in chromatography, where analyte bands are detected to allow subsequent peak integration. It is particularly challenging for comprehensive two-dimensional chromatography, due to the significant discrepancy between the first -and second-dimension information density. In this work, the effect of different peak characteristics on peak detection in comprehensive two-dimensional chromatography was investigated using simulated data based on probability distributions originating from experimental data.

Two fundamentally different peak detection strategies were studied: the two-step and watershed approaches. A "stationary" peak was placed at the center of a 100x100 grid, and a second "moving" peak was placed along each of the 10,000 grid positions. For each position, the recovery of the peak area was calculated as a function of different peak characteristics: peak width in both the first and second dimension, relative peak ratios, peak asymmetry, peak shape, and two types of modulation shifting. These seven parameters, with 70 total values across all of them, were each varied at different levels and degrees of resolution. Overall, this resulted in a total of 700,000 chromatograms for which the peak detection was evaluated for both peak detection algorithms which gave insight into the effects of peak characteristics on the performance of the peak detection algorithmns. This analysis provided insights into how peak characteristics influence the performance of the peak detection algorithms.

This work allows the evaluation of signal-processing algorithms with a level of objectiveness, systematism, and scale that would be unachievable using experimental data. Assessment of strategies (i) allows for a more informed decision about the selection of a suitable peak-detection strategy, (ii) provides insights that can be used to advance peak detection through new algorithms that address the weaknesses, and (iii) provides an objective way to benchmark any existing and future peak detection method.