Conductance-based dendrites perform Bayes-optimal cue integration.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Theoretical Neuroscience

Background:

• Cortical circuits integrate diverse information for behavior, often mirroring optimal Bayesian probability theory.
• The biological mechanisms underlying this optimal information integration in neural substrates remain largely unknown.
• Existing models lack a clear link between neural dynamics and Bayesian computation.

Purpose of the Study:

• To propose a novel Bayesian computational framework for understanding neural information integration.
• To elucidate how conductance-based neurons and synapses can naturally perform Bayesian computations.
• To derive a synaptic plasticity rule consistent with Bayesian principles.

Main Methods:

• Developed a theoretical model viewing neuronal compartments (apical and basal dendrites) as representing Bayesian priors and likelihoods.
• Formally demonstrated how somatic integration computes posterior probabilities based on dendritic inputs.
• Derived a gradient-based synaptic plasticity rule for learning distributions and weighting inputs by reliability.

Main Results:

• Showed that apical dendrites encode prior expectations and basal dendrites encode likelihoods.
• Demonstrated that somatic compartments naturally compute posterior probabilities.
• Derived a plasticity rule enabling neurons to learn distributions and adapt synaptic weights based on input reliability.

Conclusions:

• The proposed Bayesian view offers a biologically plausible mechanism for optimal information integration in cortical circuits.
• The model successfully explains existing experimental findings on multi-sensory integration at system and single-cell levels.
• The theory yields testable predictions for Bayesian dendritic integration and synaptic plasticity.