Using spike train distances to identify the most discriminative neuronal subpopulation.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Data Analysis

Background:

• Neuronal population responses can be analyzed under summed population (SP) or labeled line (LL) hypotheses.
• External stimuli evoke responses from the entire neuronal population or individual neurons independently.
• SPIKE-distance quantifies these responses, either pooled across a population or per neuron.

Purpose of the Study:

• To develop and compare algorithms for analyzing neural population coding under SP and LL hypotheses.
• To introduce novel methods for identifying discriminative subpopulations and combining individual neuron contributions.
• To provide tools for quantifying neuronal responses in diverse coding scenarios.

Main Methods:

• For SP coding, three algorithms were compared for identifying the most discriminative subpopulation across stimulus pairs.
• A new algorithm was developed for LL coding, combining neurons that best distinguish stimulus pairs individually.
• Simulated annealing was found to outperform gradient algorithms for SP analysis in realistic population sizes.

Main Results:

• Brute force search is optimal for SP but only feasible for small populations.
• Simulated annealing offers a practical and effective alternative for SP analysis in larger populations.
• The novel LL algorithm efficiently handles complex coding scenarios.

Conclusions:

• The developed algorithms correctly identify neuronal responses across the spectrum of coding possibilities, from SP to LL extremes.
• These methods offer a complementary approach to existing spike train distance analyses by focusing explicitly on subpopulations.
• The findings advance the understanding of neural population coding and provide practical tools for analysis.