Indistinguishability domains of neural microcircuit motifs mapped through classification scores of postsynaptic spike counts

In the kinetic modeling of a chemical synapse, the reversal potential ( E _syn ) and conductance amplitude of the synapse jointly regulate the spike-induced activity of the postsynaptic neuron. We simulate two-neuron microcircuit motifs of feedforward ( i → j ) and feedback ( i ⇌ j ) types using random external current pulses provided only at the presynaptic neuron. For combinations of E _syn and corresponding to an excitatory synapse, a supervised machine learning method for classification distinguishes the motifs perfectly with a score of 1.0 when using binned counts of postsynaptic spikes as the input. Strongly inhibitory combinations of these parameters result in no postsynaptic response in both types of microcircuit motifs; hence, the classifier does not improve upon a random assignment with a score of 0.5. In other domains of the parameter space spanned by E _syn and corresponding to diminished excitation/inhibition, the classifier fails (0.5 < score < 1), indicating the challenge in identifying the nature of the synapse inferred exclusively from postsynaptic spikes. For this task, the gradient boosting classifier gives higher classification accuracies that improve with further training, compared to other classifiers explored in this study.