Fig. 3: Topology estimation of Markov models using neural networks.
From: A deep learning approach to real-time Markov modeling of ion channel gating
A Shows all possible linear five-state topologies that were all encompassed in the training dataset for the topology estimation. They are grouped according to the number of open/closed states and their interconductance rank (number of independent C-O links). B The accuracy related to the size of the training dataset is displayed. The NN (Fig. 1A) was trained using subsets of dataset No. 1 (Table 1) with varying training dataset sizes. The training was repeated three times for each dataset, and the average, together with the standard deviation, is depicted. For the training dataset size of 107, the patience for the learning rate reduction and training termination was reduced to 4 and 6, respectively. C, D The confusion matrices were obtained by testing a single NN that has been trained with 107 2D-histograms. The axes show the index of the topology, which can be gathered from (A). C The recall (diagonal values) and the False Negative Rate (FNR) (off-diagonal values) are displayed. D The precision (diagonal values) and the False Discovery Rate (FDR) (off-diagonal values) are displayed (see methods).
