Understanding how learning and memory formation work in the brain is a major challenge in neuroscience, with important implications for many other fields, including medicine and industry. It is nowadays widely accepted that synaptic plasticity is the biological foundation of these higher order brain functions. So far, many different plastic behaviors have been intensively studied and characterized, leading to the definition of several forms of plasticity: structural, functional, homeostatic, inhibitory, and many others. Unfortunately, despite all the interest and efforts of the scientific community, a complete and consistent understanding of synaptic plasticity is still lacking.
The main goal of this study is to unify data and theories on synaptic plasticity in a comprehensive model, suitable for studying learning and memory down to the synapse level. To reach our objective, we identified a minimal set of biological mechanisms responsible for plastic dynamics and integrated them into a single synapse model, relying whenever possible on well accepted sub-models from literature. We designed a data-driven fitting and generalization strategy to parameterize all excitatory-to-excitatory synapses in a large scale reconstruction of neocortical tissue [Markram et al., 2015]. Finally, we tested the effects of functional synaptic plasticity on neural circuits in simulations.
Our model was able to capture not only the outcome of Spike Timing Dependent Plasticity (STDP) protocols used in the training phase, but also the one of all others available in the same experimental dataset [Markram et al., 1997b]. Moreover, it correctly reproduced results on distance-dependent synaptic plasticity [Sjöström and Häusser, 2006], even though this dataset was never used during fitting. In network simulations, we observed the emergence of a self-regulatory homeostatic mechanism, preventing runaway excitation. Furthermore, we noticed the strengthening of connections between neurons that are similarly innervated, as previously shown in vitro [Perin et al., 2011].