Normative and mechanistic model of an adaptive circuit for efficient encoding and feature extraction

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Relating neural connectomes to activity, circuit function, and learning remains a key challenge.
• The peripheral olfactory circuit of Drosophila larva, with its receptor neurons and local interneurons, offers a model system to address this.
• Understanding feedback loops and inhibitory local neurons is crucial for deciphering circuit dynamics.

Purpose of the Study:

• To develop biologically plausible mechanistic models of the Drosophila larval olfactory circuit.
• To investigate how connectomes relate to neural activity, circuit function, and learning.
• To propose a general circuit motif for efficient sensory information processing and adaptation.

Main Methods:

• Combined structural and activity data from the Drosophila larval olfactory circuit.
• Employed a holistic normative framework based on similarity-matching.
• Formulated and analyzed linear and nonnegative circuit models.

Main Results:

• The nonnegative model predicted observed synaptic weights and revealed they reflect ORN activity correlations.
• The model explained relationships between synaptic counts and the emergence of different local neuron types.
• Proposed that local neurons encode activity cluster memberships, whiten, and normalize ORN representations via feedback.

Conclusions:

• The identified circuit motif can autonomously emerge through Hebbian plasticity for unsupervised adaptation.
• This motif efficiently extracts input features and optimizes neural representations.
• The study provides a unified framework linking structure, activity, function, and learning, supporting similarity-matching's role in neural representation transformation.