Comparison of derivative-based and correlation-based methods to estimate effective connectivity in neural networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Biology

Background:

• Understanding neural system connectivity from activity patterns is crucial but challenging.
• No universally accepted method exists for inferring directed effective connectivity.
• Accurate connectivity inference is vital for comprehending neural network function.

Purpose of the Study:

• To systematically compare different methods for estimating effective connectivity in neural networks.
• To identify the most reliable approach for reconstructing connectivity from observed activity.
• To provide practical guidance for effective connectivity estimation in biological research.

Main Methods:

• Utilized the Hopf neuron model with known ground truth connectivity.
• Reconstructed connectivity matrices using various algorithms, including lagged-cross-correlation (LCC) and derivative-based covariance analysis.
• Compared method performance in purely excitatory, sparse, non-linear networks with delays and noise, as well as linear networks.

Main Results:

• A combined LCC and derivative-based covariance analysis method showed the most reliable estimation of ground truth connectivity in sparse non-linear networks with delays.
• LCC demonstrated comparable performance to transfer entropy in linear networks but with significantly lower computational cost.
• The LCC method performed better than a reservoir computing-based method on empirical data from *C. elegans*.

Conclusions:

• A computationally simple LCC-based method reliably estimates directed effective connectivity in sparse neural systems with spatio-temporal delays and noise.
• The findings offer concrete suggestions for effective connectivity estimation when only neuronal activity is known.
• This approach is valuable for analyzing complex neural systems in biological research.