Approximate Bayesian computation (ABC) is a popular tool for inferring stochastic reaction networks from single-cell data, but it typically relies on extensive Gillespie simulations that are computationally expensive. A new preprint by Zekai Li, Mauricio Barahona, and Philipp Thomas proposes a simulation-free alternative called SFABC, which avoids both simulation and explicit likelihood solutions of the network dynamics.

Instead of simulating trajectories, SFABC uses the Chemical Master Equation to set up a tractable convex optimization problem with constraints. These constraints are then used to accept sampled parameters that are consistent with the observed data. The authors prove that the resulting posterior approximation can be tightened systematically and converges to the exact Bayesian posterior distribution.

In synthetic benchmarks, SFABC matches the accuracy of standard simulation-based algorithms while running at substantially reduced computational cost. The method is also demonstrated on experimental single-cell data, where it analyzes gene transcription bursting kinetics from snapshot, time-resolved, and cell-cycle-resolved measurements.