Figure 10.

Simulations using spike trains with shuffled ISIs. Temporal pattern of pre-synaptic firing rate and calcium dynamics are different with shuffled spike trains, but still predict weight change. (A,B) The end weight of one set of 25 simulations using 208 spike trains with shuffled ISIs was subdivided into seven weight change bins. (A) Weight change triggered average pre-synaptic firing rate distinguishes weight change bins but differs from the temporal pattern of non-shuffled spike trains. (B) Weight change triggered average calcium concentration exhibits temporal pattern similar to non-shuffled spike trains, with higher values later in the trial for synapses that potentiate. (C–E) For each of 5 sets of 25 simulations, we used 75% of the data as the training set and the remainder as test set, repeated 4 times, giving a total of 20 regressions for each combination of features. (C) Random Forest regression using calcium time samples as inputs is better with 3 or 5 time samples compared to 1 time sample. (D) Feature importance from random forest regression using calcium time samples as inputs shows that the middle samples are more important than the ending sample. (E) Random Forest regression using time samples of pre-synaptic firing and spatial feature as inputs shows that using 3 time samples has better prediction score than 1 time sample. Cluster length improves the prediction score independent of the number of time samples.