Global and Multiplexed Dendritic Computations under In Vivo-like Conditions.

Dendrites integrate inputs in highly non-linear ways, but it is unclear how these non-linearities contribute to the overall input-output transformation of single neurons. Here, we developed statistically principled methods using a hierarchical cascade of linear-nonlinear subunits (hLN) to model the dynamically evolving somatic response of neurons receiving complex spatio-temporal synaptic input patterns. We used the hLN to predict the membrane potential of a detailed biophysical model of a L2/3 pyramidal cell receiving in vivo-like synaptic input and reproducing in vivo dendritic recordings. We found that more than 90% of the somatic response could be captured by linear integration followed a single global non-linearity. Multiplexing inputs into parallel processing channels could improve prediction accuracy by as much as additional layers of local non-linearities. These results provide a data-driven characterisation of a key building block of cortical circuit computations: dendritic integration and the input-output transformation of single neurons during in vivo-like conditions.