Fig. 4: Comparison of the RB-CME solver and Monte Carlo method in the linear network with six species.
From: Advanced methods for gene network identification and noise decomposition from single-cell data
A The diagram of the linear network with six species: all the settings are the same as that in Fig. 3. B Convergence of both approaches in terms of the L1 error. Both methods converge at the rate of \(1/\sqrt{N}\) or \(1/\sqrt{T}\), where N and T are the sample size and computational time, respectively. This also implies that for both approaches, the computational time is proportional to the sample size (see the last block). With the same sample size, the RB-CME solver is 20 times more accurate than the Monte Carlo method, and at the same time cost, the RB-CME solver is 8 times more accurate. C Performance of the RB-CME solver (with 104 samples) and Monte Carlo method (with 105 samples) in estimating the marginal distributions. Both methods accurately approximate the marginal distributions for individual species, but they perform quite differently in estimating the joint probabilities. Specifically, the RB-CME solver is more accurate in estimating the follower system, particularly the first follower subsystem consisting of S1 and S2, but it is less accurate in estimating the leader system. Source data are provided as a Source Data file.
