OT-MESH is an unsupervised computational framework that utilizes entropy-regulated optimal transport (OT) to systematically determine evolutionary homologies of cell types across species. To overcome the limitations of conventional reference-based or projection-based matching methods, OT-MESH improves the OT scheme by incorporating the Minimize Entropy of Sinkhorn (MESH) technique, transforming diffusion transport matrices into interpretable sparse correspondences. Evaluations on synthetic datasets demonstrate high accuracy, computational efficiency, and robustness to noise. Compared to other OT-based methods, such as RefCM, OT-MESH achieves comparable accuracy with improved speed. Applications to mouse and macaque retinal bipolar cells (BCs) and retinal ganglion cells (RGCs) accurately recover known evolutionary relationships and discover novel correspondences, one of which is validated in independent experiments. Thus, OT-MESH provides a principled, scalable, and interpretable evolutionary cell type mapping solution that provides deep insights into cell specialization and conservation across species.
