Ensemble stacking mitigates biases in inference of synaptic connectivity

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Network Science

Background:

• Directly measuring neuronal anatomical connections is challenging.
• Inferring connections from neuronal activity offers a promising alternative.
• Statistical regularities in spike timing are used to infer excitatory synaptic connections.

Purpose of the Study:

• Compare and contrast commonly used inference methods for neuronal connectivity.
• Identify strategies to improve the accuracy and reduce biases of these methods.
• Develop an ensemble approach to enhance the reliability of connection inference.

Main Methods:

• Utilized simulated spiking neuronal networks to test inference algorithms.
• Applied mutual-information-based and frequency-based methods for connection inference.
• Developed and evaluated an ensemble prediction strategy combining multiple algorithms.

Main Results:

• Simple adjustments, like a signing procedure and a correction for background timing, improve inference accuracy.
• Different inference methods reveal distinct network subsets and exhibit unique biases.
• Ensemble predictions demonstrated higher sensitivity and faithfulness to ground-truth connectivity compared to individual methods.

Conclusions:

• Adjustments to standard algorithms enhance neuronal connection inference.
• Ensemble approaches mitigate biases and improve accuracy in mapping synaptic networks.
• Ensemble-based methods show broad utility and potential for future neuroscience research.