Regime-Aware Trajectory Forecasting

RegimeFlow

One generative model that forecasts the future dynamics of 1,050 biological systems — and generalizes to systems it has never seen. Instead of hand-crafting a new ODE for every system, RegimeFlow learns the shared “grammar” of biological dynamics from simulated trajectories, and predicts future states.

At a glance

1,050ODE models from BioModels
−31%MAE vs. baselines
+17%CRPS · uncertainty calibration
96 → 256context → forecast horizon

Benchmark comparison

MSE comparison between Ours and five competitive baselines across ten most frequent biological system taxonomies.

Point Forecasting Zero-Shot TSFM Probabilistic w/o Regime Probabilistic w/ Regime

Overall MSE across all models (lower is better).

From mechanism models to a universal predictor

🧬

BioModels

1,050 curated ODE mechanism models, 1–1,000 species each.

📊

SysBio-Traj

Numerically solved trajectories, 512 time points per system.

🎯

RegimeFlow

Regime-aware flow matching, trained once across all systems.

🌐

Web Demo

Past 96 points → future 256 points, with uncertainty.

Four core ingredients

🏷️

Regime Classification

Stable / oscillatory / monotonic behavior, injected as conditioning signal C.

🎛️

Regime-Aware Prior

Bayesian linear regression with per-regime basis functions replaces standard Gaussian noise.

➡️

Conditional Flow Matching

Learns a vector field that pushes prior samples to target trajectories.

Mamba + AdaLN

Linear-complexity state-space backbone with adaptive layer norm.

Regimes it captures

📉StableSystems that settle to a steady state
🌊OscillationSustained or damped periodic dynamics
📈MonotonicPersistent growth or decay

Why it matters

Biological systems are wildly diverse, yet their large-scale behavior collapses to a handful of regimes. By conditioning on regime and seeding generation with a regime-aware prior instead of Gaussian noise, RegimeFlow shortens the transport distance the model must learn — yielding 31% lower MAE and 17% better-calibrated uncertainty (CRPS) than baselines, while generalizing to unseen systems with no new equations and no retraining.

Taxonomic Composition

1,050 Biological Models

Click a circle to zoom in · Click background to zoom out

Systems Biology Trajectory Prediction