The relationship between biophysics and synaptic integration in cortical neurons
1 Max-Plack-Institute for Neurobiology of Behavior
2 Janelia Research Campus
3 Max-Planck-Institute for Neurobiology of Bahavior
4 Max-Planck-Institute for Neurobiology of Behavior
The dendrites of a single postsynaptic cell receive synaptic input from thousands of other presynaptic neurons, which the neuron then processes and transmits as action potentials to subsequent neurons in the network.
Different neurons do not only differ in their morphology but also in their biophysics. The differences in biophysics include ion-channel distributions as well as intracellular dynamics.
These factors enable complex, non-linear synaptic integration. However, how exactly the biophysics of different neurons determines the integration of the signals they receive remains an open question.
To explore this connection, we developed an explainable AI approach that predicts the dendritic voltage traces based on synaptic activation patterns and biophysics. Explainability is achieved by designing the ANN architecture such that one layer is a sensitivity matrix weighing the importance of synaptic input across space and time depending on biophysics. We apply these methods to cortical layer 5 pyramidal tract neurons of the rat somatosensory cortex which receive highly heterogeneous input and are diverse in their biophysical features.
Currently, the network is able to predict subthreshold dendritic voltage traces in response to synaptic activations for diverse biophysical configurations with an accuracy of 94% (Pearson R2). The explainable ANN architecture allows us to investigate how biophysical differences are reflected in spatiotemporal sensitivity to synaptic input.
Thus, the network successfully captures the relationship between biophysics and subthreshold synaptic integration. Therefore the ANN is a promising candidate for further analysis, including suprathreshold activity and additional explainability features.