Imported from hiyenwong/ai_collection (collection/skills/neuroscience/common-synaptic-input-estimation-hdemg/SKILL.md). Install upstream with npx skills add hiyenwong/ai_collection --skill common-synaptic-input-estimation-hdemg. Copyright stays with the author.

Common Synaptic Input Estimation from Motor Unit Spike Trains

Overview

Common synaptic input (CSI) represents the dominant component of neural drive transmitted from motor neurons to muscle. This methodology provides a practical, physiologically grounded guide for estimating CSI from populations of motor unit (MU) spike trains using openhdemg, an open-source Python framework.

Three Complementary Approaches

1. Time-Domain: Smoothed Discharge Rates

• Principle: Apply smoothing to instantaneous discharge rates, then compute correlation
• Physiological Interpretation: Captures shared low-frequency fluctuations in motor neuron drive
• Key Parameters: Smoothing window width (critical choice), epoch duration
• Implementation: openhdemg.csi.time_domain()

2. Frequency-Domain: Cumulative Spike Train Coherence

• Principle: Compute coherence between cumulative spike trains (CST)
• Physiological Interpretation: Identifies frequency-specific common oscillatory inputs
• Key Parameters: Segment length, overlap, frequency bands of interest
• Implementation: openhdemg.csi.coherence()
• Bands: Typically 0-5 Hz (common drive), 5-20 Hz (beta), 20-40 Hz (gamma)

3. Network-Information: Pairwise Dependencies + Graph Theory

• Principle: Nonlinear pairwise dependencies modeled as graph edges
• Physiological Interpretation: Reveals network topology of shared inputs
• Key Parameters: Dependency measure (mutual information, transfer entropy), graph threshold
• Implementation: openhdemg.csi.network()

Methodology Workflow

Step 1: HDsEMG Decomposition

import openhdemg

# Load HDsEMG data
data = openhdemg.load("emg_recording.h5")

# Decompose into motor unit spike trains
mus = openhdemg.decompose(data, method="fastica")
# Returns: list of MotorUnit objects with spike_trains attribute

Step 2: Motor Unit Cleaning

# Filter MUs by quality metrics
mus_clean = openhdemg.filter_mu(mus, 
    min_discharges=50,          # Minimum spike count
    maxisi_ratio=3.0,           # Maximum ISI ratio (refractory check)
    min_sil=0.9)                # Minimum silhouette score for decomposition quality

Step 3: CSI Estimation

Time-Domain Method

# Smoothed discharge rate correlation
csi_td = openhdemg.csi.time_domain(
    spike_trains=[mu.spike_train for mu in mus_clean],
    smoothing_window=0.4,  # seconds (critical parameter)
    bin_size=0.001,         # 1ms bins
    epoch_duration=5.0      # seconds per epoch
)
# Returns: correlation matrix (N_MUs x N_MUs)

Frequency-Domain Method

# Cumulative spike train coherence
csi_freq = openhdemg.csi.coherence(
    spike_trains=[mu.spike_train for mu in mus_clean],
    fs=1000,                # sampling frequency
    n_segments=10,          # number of segments for averaging
    overlap=0.5,            # 50% overlap
    f_range=(0, 50)         # frequency range of interest
)
# Returns: coherence spectrum, common_drive_strength per frequency band

Network-Information Method

# Pairwise nonlinear dependencies as graph
csi_net = openhdemg.csi.network(
    spike_trains=[mu.spike_train for mu in mus_clean],
    method="mutual_information",  # or "transfer_entropy"
    threshold=0.1,           # graph edge threshold
    bin_size=0.01            # discretization bin
)
# Returns: graph object with nodes=MUs, edges=shared input strength

Critical Parameter Selection Guide

Key Findings from Literature

1. Decomposition Quality Directly Affects Estimates: Poor MU decomposition propagates errors into CSI estimates
2. Time-Domain is Most Robust: For general common drive assessment
3. Frequency-Domain Reveals Oscillatory Structure: Essential for studying beta/gamma oscillations
4. Network Method Reveals Topology: Shows which MUs share inputs (not just strength)
5. No Single Method is Best: Use complementary approaches for complete picture

Pitfalls and Warnings

1. Smoothing Window: Most critical parameter in time-domain method. Default values may not suit all datasets. Always perform sensitivity analysis.
2. Number of MUs: Methods require sufficient MU population (>20 MUs recommended for stable estimates)
3. Non-stationarity: CSI can change during fatiguing contractions. Use sliding windows for time-varying analysis.
4. Crosstalk: Ensure decomposition is from single muscle; crosstalk contaminates CSI estimates.
5. Discharge Rate Confound: Common input estimates correlate with mean discharge rate. Control for this confound.

Applications

• Motor Control Research: Quantify shared neural drive during different tasks
• Clinical Assessment: Detect changes in common input in neurological disorders
• Aging Studies: Track changes in synaptic input organization with age
• Fatigue Research: Monitor how common input evolves during sustained contractions
• BCI/Neuroprosthetics: Decode motor intent from shared input patterns

Reference

Cabral, H.V., Valli, G., Zanotti, R., Delis, I., & Negro, F. (2026). Estimating common synaptic inputs to spinal motor neurons from motor unit spike trains using openhdemg. arXiv:2606.23066v1

Related Skills

• eeg-brain-connectivity-bci: Brain connectivity for BCI applications
• bci-rehabilitation-protocols: BCI rehabilitation protocols
• eeg-preprocessing-reliability: EEG preprocessing reliability assessment