Trajectory Inference for Multi-Omics Data Using Ordered Labels

Trajectory inference (TI) has emerged as a crucial approach for understanding cellular development and differentiation processes. By reconstructing the temporal ordering of individual cells, TI can analyze the mechanisms underlying development and disease progression. Various TI methods have been proposed and applied across diverse fields, including cancer research and immunology. However, most existing methods rely solely on gene expression data and do not incorporate other omics information such as DNA methylation, chromatin accessibility, and protein expression. To address this limitation, approaches that integrate multiple omics datasets for TI have been proposed. Nevertheless, these methods typically assume the use of only common variables across datasets, making it difficult to align heterogeneous data with differing measurement targets across omics layers. This limitation can lead to information loss and challenges in capturing interactions between different omics datasets, which may reduce integration accuracy. In this study, we propose a novel TI method that captures interactions among variables across different omics layers and accommodates unpaired datasets with non-overlapping measurement targets. By effectively capturing relationships between different omics layers, our approach enables a more comprehensive multi-omics TI framework, thus providing deeper insights into cellular development.