Liquid chromatography-mass spectrometry (LC-MS)-based multiomics studies require accurate cross-sample quantification and comparison of analyte abundances, but due to retention time (RT) shifts between runs, dedicated RT alignment algorithms are required to match features corresponding to the same analyte across runs. However, existing algorithms focus mainly on the superficial RT shift distances between runs, neglecting the underlying chromatographic principles that govern these shifts. Here, we introduce Trend-Aligner, the first feature alignment algorithm for untargeted MS data that explicitly models RT shifts based on chromatographic shift principles to achieve fine-grained, analyte-wise RT correction across runs. Trend-Aligner decomposes RT shifts into global and local components. The global RT shift is described using a nonlinear warping function that captures systematic variations in chromatographic conditions across runs, while the local RT shift is modeled using a latent factor model, capturing how analytes with different physicochemical properties respond differently to changes in chromatographic conditions. To comprehensively evaluate performance, we developed a reference-based accuracy benchmarking strategy and manually annotated four metabolomic and five proteomic data sets as reference sets, comprising 3663 consensus features and 57,652 feature peaks. Compared to 11 widely used alignment algorithms, Trend-Aligner consistently demonstrated the highest accuracy across data sets. To assess real-world applicability, we further conducted application-oriented utility validation by evaluating both the quantity and quality of aligned features when integrated with the match-between-runs (MBR) function. Trend-Aligner delivered the best performance, combining high sensitivity and specificity, and exhibited a remarkable 82.5% increase in post-MBR identified peptides compared to MaxQuant. Trend-Aligner is open-source and freely available at https://github.com/CSi-Ti/trend-aligner.