Noise‑robust moving window ANOVA algorithm for accurate peak detection and fraction collection in preparative liquid chromatography.

Preparative liquid chromatography (prep-LC) is essential for purifying natural products and pharmaceuticals, but accurate real-time peak detection remains challenging under high-noise, high-flow-rate conditions. Conventional threshold- and derivative-based algorithms often fail to detect shallow, slowly rising peaks because signal changes are masked by baseline fluctuations, and smoothing preprocesses can blur genuine peak features. To overcome these limitations, we developed a moving-window analysis of variance (MW-ANOVA) algorithm that performs an F-test on two consecutive subgroups within a sliding window to distinguish systematic chromatographic trends from random noise. The method requires no peak-shape assumptions or prior denoising. We evaluated MW-ANOVA on analytical-scale symmetric standards, on preparative-scale separation of Houttuynia cordata extract, and on medium-pressure liquid chromatography (MPLC) purification of Lonicera japonica. In the preparative separation, MW-ANOVA reliably detected shallow peaks that were missed by the first derivative method, with a minimum detectable slope of approximately 0.07 mV per data point. It guided automated collection of 28 fractions, three of which showed area percentage purities above 93 % at 254 nm. In the MPLC run, bubble peaks and broad elution profiles were present; MW-ANOVA correctly identified the true elution peaks and ignored bubble peaks, whereas the first-derivative method produced 159 mostly false-positive peaks and the fixed-threshold method failed to define correct boundaries. The algorithm was also robust against gradient-induced baseline drift. MW-ANOVA provides a practical, statistically grounded solution for peak detection in preparative and medium-pressure LC, especially for broad, asymmetric, and low-amplitude peaks.