Navigo

Navigo is a biologically grounded generative modeling framework for developmental biology, designed to reconstruct continuous cell-fate trajectories from temporal single-cell RNA sequencing (scRNA-seq) snapshots. It was developed by the aristoteleo lab (Yimin Fan, Xinyuan Liu, Yu Li and colleagues at The Chinese University of Hong Kong, with Xiaojie Qiu and Zehua Zeng at Stanford University School of Medicine) and released as a bioRxiv preprint in June 2026. The group is the same team behind Dynamo, and Navigo extends that lineage of RNA-velocity and vector-field methods into the generative-modeling era.

The central problem Navigo addresses is that single-cell atlases capture development only as static snapshots: cells are destroyed at the moment they are measured, so the continuous dynamics connecting early progenitors to differentiated states must be inferred. Navigo learns a developmental vector field by combining flow matching at the population level — which transports the distribution of cells smoothly across time — with RNA kinetics modeling at the molecular level, which grounds those transitions in the transcription, splicing, and degradation processes that actually drive gene-expression change.

By unifying these two scales, Navigo positions itself not merely as a trajectory-interpolation tool but as a generalizable platform for perturbation-effect prediction, disease modeling, and rational cell-fate engineering, learned from a single large reference atlas rather than retrained per task.

#Key Features

• Two-scale vector field: Couples population-level flow matching with molecular-level RNA kinetics, so the learned developmental dynamics are both distributionally accurate and mechanistically interpretable.
• Zero-shot perturbation prediction: Predicts the effects of genetic perturbations without perturbation-specific training, validated against six knockout (KO) genotypes.
• Disease modeling: Maps regulatory dysregulation underlying developmental disease, demonstrated on congenital heart disease subtypes.
• Cell-fate engineering: Screens transcription-factor combinations to guide reprogramming interventions, illustrated through fibroblast reprogramming.
• Atlas-scale trajectory mapping: Reconstructs and denoises developmental trajectories across lineages on a 12.4-million-cell atlas spanning 43 developmental time points.

#Technical Details

Navigo formulates development as a continuous vector field learned via flow matching, a simulation-free generative training objective that fits a velocity field transporting cells between consecutive temporal distributions. This population-level transport is regularized by an RNA kinetics model operating at the molecular level, linking observed expression changes to underlying transcriptional and splicing dynamics rather than treating the field as a purely geometric interpolation. The framework was trained on a mouse embryogenesis scRNA-seq atlas of approximately 12.4 million cells across 43 time points, providing dense temporal coverage of lineage diversification. The released implementation (Python, BSD-2-Clause licensed) ships tutorials for training, trajectory interpolation and denoising, gene regulatory network (GRN) analysis, knockout studies, and reprogramming workflows. Pretrained checkpoints are distributed as downloadable tutorial asset bundles referenced in the documentation rather than bundled in the repository.

#Applications

Navigo is aimed at developmental biologists, stem-cell researchers, and computational biology groups studying how cell fates are specified and how they go awry. Its zero-shot perturbation prediction lets researchers prioritize candidate regulators in silico before committing to costly knockout or perturbation experiments, while its disease-modeling capability — shown for congenital heart disease subtypes — supports hypothesis generation about the regulatory origins of developmental disorders. The cell-fate engineering workflow, demonstrated on fibroblast reprogramming, offers a principled way to nominate transcription-factor combinations for directed differentiation and reprogramming, integrating into existing single-cell analysis pipelines that already produce time-resolved scRNA-seq data.

#Impact

By bringing flow matching together with mechanistic RNA kinetics at atlas scale, Navigo advances single-cell developmental modeling beyond descriptive trajectory inference toward a predictive, intervention-oriented framework. Coming from the aristoteleo lab, whose Dynamo toolkit is widely used for RNA-velocity vector-field analysis, it is positioned to influence how the field models continuous developmental dynamics and perturbation responses. As a recent preprint, its broader adoption and independent benchmarking remain to be established. The preprint is distributed under a CC-BY-NC-ND license and the code under BSD-2-Clause; the pretrained checkpoints are released without a separately stated license.