DualRTWarp: Robust Correction of Retention Time Shifts in Chromatography-Mass Spectrometry Data——Halo96 and AOA74 datasets

The Ha96 and AOA74 lipidomic dataset is used to benchmark the performance of DualRTWarp, a two-step RT correction algorithm designed to handle datasets with complex and substantial RT shifts. The Ha96 lipidomic dataset includes 77 total lipid extract (TLE) samples from halobacterial pure cultures, 8 pooled samples prepared from TLEs, and 11 blank samples interspersed throughout the analytical batches. This dataset includes seven strains representing three distinct taxonomic orders, exhibiting substantial variation in lipidome composition (Yao et al., 2023): Haloarcula argentinensis, Haloferax larsenii, Haloferax mediterranei, Haloferax volcanii, Haloterrigena turkmenica, Natrialba asiatica, and Natrinema gari. The TLEs were extracted using a modified Bligh and Dyer method (Bligh & Dyer, 1959; Findlay, 2004), analyzed using a Waters ACQUITY I-class ultra-performance liquid chromatography coupled to Waters SYNAPT G2-Si quadrupole time-of-flight mass spectrometer using reversed phase chromatography (Yao et al., 2023). Most samples were analyzed using a Waters Acquity BEH C18 column (150 × 2.1 mm, 1.7 μm particle size) across two batches spaced two months apart in 2022, and one mixed TLE sample was tested in 2021. The 45-minute chromatographic runs exhibited complex patterns of RT shifts, with shifts up to 1.85 minutes. The AOA74 lipidomic dataset includes 66 TLE samples from ammonia-oxidizing archaea (AOA) cultures and 8 solvent blanks. TLE were extracted using a modified Bligh and Dyer method (Bligh & Dyer, 1959). The data was analyzed using a Vanquish Horizon UHPLC system equipped with a Waters Acquity BEH C18 column (150 × 2.1 mm, 1.7 μm particle size) coupled to a Q Exactive Plus Orbitrap high-resolution MS (Thermo Fisher Scientific, Bremen, Germany) equipped with a heated electrospray ionization source (HESI) operating in positive ionization mode. Reversed-phase chromatographic conditions were as described previously (Wörmer et al., 2013). The datasets were analyzed in two batches spaced 1 year apart, resulting in an inconsistent RT pattern.