Fig. 4

Genome-wide optimization of CTCF binding kinetics underlying loop extrusion. A A Markov chain with an Unbound (red) and Bound (blue) state underlies MoDLE loop extrusion barrier modeling. The self-transition probability for the Bound state (P_BB) reflects how stably barrier elements (i.e., CTCF) are bound to their binding sites. The stationary distribution of the Markov chain (π_B) provides the CTCF binding probability at a given epoch in the simulation. The bottom diagram (red/blue boxes) shows an illustration of how the binding state (Bound in blue, and Unbound in red) of a single CTCF site would change during a simulation depending on P_UU and P_BB. B Convergence of the objective function during the Bayesian optimization procedure. The objective function is a dissimilarity score comparing the pixels showing stripes and dots in the observed Micro-C data with the corresponding stripes and dots in the MoDLE output. See the “Methods” section (part 6) for details. C Comparison of objective function in the parameter search space of P_UU and π_B. Optimal, near-optimal, suboptimal and non-optimal combinations are highlighted with a red star, orange pentagon, blue square and green triangle respectively. D Side-by-side comparison of H1-hESC Micro-C data (top panel) and progressively less optimal combinations of P_UU and π_B parameters