RT Journal Article SR Electronic T1 Non-linear dendrites enable robust stimulus selectivity JF bioRxiv FD Cold Spring Harbor Laboratory SP 023200 DO 10.1101/023200 A1 Romain D. Cazé A1 Sarah Jarvis A1 Simon R. Schultz YR 2015 UL http://biorxiv.org/content/early/2015/12/12/023200.abstract AB Hearing, vision, touch – underlying all of these senses is stimulus selectivity, a robust information processing operation in which cortical neurons respond more to some stimuli than to others. For vision, Hubel and Wiesel discovered that certain neurons respond selectively to elongated visual stimuli, and proposed an elementary linear model to account for this selectivity. Recent experiments have however cast doubt on some aspects of this textbook model. Hyperpolarising a neuron can abolish selectivity in the soma, while selectivity in the dendrites - the receptive compartments of neurons - remains unaffected. A model assuming linear summation of inputs, like the Hubel and Wiesel model, cannot explain this observation. Here instead we employ a morphologically realistic model, incorporating non-linear dendrites reconstructed from real neurons, to implement stimulus selectivity. We show that this model explains the effect of hyperpolarisation, and implements stimulus selectivity more robustly than the classic model. It can remain selective even if 50% of its synapses fail. This demonstrates that in addition to increasing neuronal computational capacity, dendrites can also increase the robustness of neuronal computation. We also predict that a neuron that is initially stimulus non-selective can become selective when depolarized. This prediction stimulates new experimental studies on stimulus selectivity.Moreover, the robustness of our implementation provides a starting point for the development of fault-resistant neuromorphic chips.