synaptic-motifs-mean-field

name	synaptic-motifs-mean-field
description	Mean-field theory linking microscale synaptic motifs to macroscopic heterogeneous population dynamics. Bridges synaptic-resolution connectomics with nonlinear neural dynamics via low-rank mean-field equations. Applicable to RNN analysis, neural population modeling, V1 response prediction.
arxiv_id	2606.27946
tags	["neural-dynamics","mean-field","synaptic-motifs","connectomics","recurrent-networks","population-dynamics"]

Synaptic Motifs Mean-Field Theory

Background

Recent breakthroughs in synaptic-resolution network connectomics reveal fine-scale structural connectivity — correlated synaptic coupling pairs known as second-order motifs. Large-scale recordings show macroscopic heterogeneous population dynamics. This framework bridges the gap: can microscale synaptic structures contribute to macroscopic heterogeneous dynamics in ways canonical models cannot?

Core Methodology

Random RNNs with Structured Synaptic Statistics

Setup:

• P population RNN with various cell types
• Nonlinear non-negative neural responses
• Arbitrary marginal AND second-order correlated synaptic statistics

Key Innovation — Low-Rank Mean-Field Equations:

For P-population networks, derive equations where:

• Pre- and postsynaptic neuronal population identities → determine synaptic AND motif strengths
• Framework requires 2P latent dynamic variables:
  • P variables: mean population activity
  • P variables: within-population variability

Chain Motif Mechanism

Chain motifs induce correlations in synaptic variability → microscopic fluctuations integrate → influence mesoscopic mean population dynamics.

Critical insight: Canonical brain circuit models (which ignore second-order motifs) CANNOT reproduce this bridging effect.

Application: Reverse Engineering V1 Connectivity

Apply framework to reverse engineer network connectivity that recapitulates heterogeneous activity across populations in mouse primary visual cortex (V1).

Key Results

1. Chain motifs → synaptic variability correlations → micro-to-meso integration
2. 2P-dimensional state space captures both mean activity AND variability
3. Successfully predicts V1 heterogeneous population dynamics from synaptic statistics
4. Offers testable predictions about fine-scale connectivity → dynamics → computation relationships

Implementation Guide

# Pseudo-code for mean-field equations
def mean_field_P_populations(W_mean, W_motif, P, response_func):
    """
    W_mean: P x P mean synaptic weight matrix
    W_motif: P x P x P second-order motif tensor
    P: number of populations
    response_func: nonlinear non-negative activation
    
    Returns: 2P-dimensional ODE system
    """
    # P variables for mean activity (m_1, ..., m_P)
    # P variables for variability (v_1, ..., v_P)
    # dm_i/dt = -m_i + response_func(sum_j W_mean_ij * m_j + correction_from_motifs)
    # dv_i/dt = -v_i + contribution_from_chain_motifs(W_motif, m, v)
    pass

Pitfalls

• Do NOT ignore second-order motifs — they are the key bridging mechanism
• Non-negative neural responses are essential for the theoretical derivation
• The 2P formulation (not P) is necessary — variability variables cannot be eliminated
• Finite-size corrections matter when population sizes are small (<50 neurons)

Verification

• Compare mean-field predictions against full network simulations
• Check that chain motif contributions vanish when second-order correlations are removed
• Validate V1 predictions against experimental heterogeneous activity patterns

Activation Triggers

Keywords: synaptic motifs, mean-field, population dynamics, connectomics, second-order motifs, chain motifs, heterogeneous dynamics, V1 modeling, low-rank equations, random RNN