#4 · ML / AI methods

Recovering single-cell kinetics and trajectories from static and noisy measurements

The dominant thread is dynamical modeling that recovers unobserved temporal or kinetic structure from static or noisy single-cell and single-particle measurements. The functional KL estimator frames trajectory inference as a comparison of path-space measures, validated on a benchmark before application to scRNA-seq. HMCVelo adapts deterministic ODE velocity modeling to methylation, hydroxymethylation and demethylation dynamics per cell and gene, using Joint-snhmC-seq data. The anchored Brownian motion method addresses the analogous inference problem at single-molecule scale, estimating diffusion coefficients from live-cell tracking within a Bayesian framework. What is new is the transfer of velocity-style kinetic modeling to epigenomic modification states, alongside a divergence criterion for comparing trajectories. The works share a modeling logic rather than a common dataset.