Model selection results for ISI of synthetic PCs.

Marginal likelihood of the first five models in the family for each data set, corresponding to the six different injected currents. Error bars on the log-marginal likelihoods are in S2 Table. Last column shows that a model with 4 completion paths is optimal over the combined data. Asterisk marks those cases where the optimal parameter values fell at the boundary of the search space, usually because there were paths with near-zero flux through them (see Materials and methods). Note that the numbers in the first two columns increase monotonically with M, so that the best model in the family is not found for M ≤ 5. We chose to truncate the exploration at M = 5 since we are interested in the overall maximum of the log-likelihood for all I, which is reached at M = 4 (last column).

针对每个数据集，结合六种不同的注入电流工况，我们给出了该模型族内前五个模型的边际似然值。对数边际似然的误差棒信息详见附表S2。最后一列显示，在合并数据集上，包含4条完备路径的模型为最优模型。星号用于标记最优参数值落在搜索空间边界的情形，此类情况通常源于对应路径的通量近乎为零（详见材料与方法部分）。需注意，前两列的数值随M单调递增，因此当M≤5时无法获取该模型族中的最优模型。鉴于我们关注所有电流工况下对数似然的全局最大值（该最大值在M=4时取得，即最后一列的结果），故将探索范围截断至M=5。