[Paper Review] Synergy from Silence in a Combinatorial Neural Code

Elad Schneidman, Jason Puchalla+4 more|ArXiv.org|Jul 13, 2006

Neural dynamics and brain function38 references4 citations

TL;DR

This study reveals that in retinal neural codes, patterns of spiking and silence—particularly silent periods—generate significant synergy, where the collective activity conveys more information than the sum of individual neurons. Using information theory on multi-electrode recordings during naturalistic stimuli, the authors show that synchronous spikes are mostly redundant, while silent patterns enable unique, synergistic coding of stimulus features.

ABSTRACT

The manner in which groups of neurons represent events in the external world is fundamental to neuroscience. Here, we analyze the population code of the retina during naturalistic stimulation and show that the information conveyed by specific multi-neuronal firing patterns can be very different from the sum of the pattern's parts. Synchronous spikes convey more information than either of the participating cells, but almost always less than their sum, making them redundant coding symbols. Surprisingly, patterns of spiking and silence are mostly synergistic - carrying information that is accessible only by observing the whole pattern of activity, rather than its components, and signifying unique features in the stimulus. These results demonstrate that the retina uses a combinatorial code and that the brain can benefit significantly from recognizing multi-neuronal firing patterns.

Motivation & Objective

• To investigate how multi-neuronal firing patterns in the retina encode information beyond the sum of individual neuron contributions.
• To determine whether specific patterns of spiking and silence are synergistic, redundant, or independent in conveying stimulus information.
• To explore the functional significance of combinatorial coding in neural populations using naturalistic stimuli.
• To test whether synergy arises from silent periods rather than synchronous spikes, challenging prevailing assumptions about neural coding.

Proposed method

• Measured information carried by neural symbols (spike patterns) using a time-varying symbol rate model, defined as $ I(\sigma;S) = \frac{1}{T}\int_0^T \frac{r_\sigma(t)}{\bar{r}_\sigma} \log_2 \frac{r_\sigma(t)}{\bar{r}_\sigma} dt $, where $ r_\sigma(t) $ is the time-dependent rate of symbol occurrence.
• Used peri-stimulus time histograms (PSTHs) across repeated stimulus presentations to estimate symbol rates and their modulations by dynamic stimuli.
• Quantified potential synergy by computing the maximum possible information for a joint symbol (e.g., $1\otimes1$) under constraints of individual cell firing rates, using a greedy algorithm to optimize rate distribution across time bins.
• Defined synergy as the difference between the maximum achievable information of a joint pattern and the sum of individual cell information, providing a lower bound on actual synergy.
• Developed a generic model of ganglion cell responses using a sigmoidal function of stimulus projection onto a preferred direction vector, with added noise to simulate variability.
• Evaluated synergy and redundancy across pairs and triplets of neurons by averaging over a large ensemble of random stimuli and 100 neurons with uniformly distributed preferred directions.

Experimental results

Research questions

• RQ1 Do multi-neuronal firing patterns in the retina convey more information than the sum of their individual parts?
• RQ2 Is synergy primarily driven by synchronous spiking or by patterns involving silence?
• RQ3 How does the information content of specific neural symbols (e.g., spike pairs or silent patterns) relate to stimulus dynamics?
• RQ4 Can a generic model of neural response properties reproduce the observed synergy and redundancy patterns?
• RQ5 What functional role does combinatorial coding—especially from silent periods—play in neural representation?

Key findings

• Patterns involving silence, especially in combination with spiking, exhibit strong synergy, where the whole pattern conveys information not accessible from individual neurons.
• Synchronous spike pairs are typically redundant, conveying less information than the sum of individual contributions, despite being more informative than either cell alone.
• The maximum potential synergy for joint spiking patterns was found to be significantly higher than the actual observed synergy, indicating that synergy is constrained by biological response statistics.
• The average information conveyed by silent patterns was substantial and often exceeded that of spike pairs, highlighting their functional importance in neural coding.
• A generic model of neural responses with sigmoidal tuning and noise successfully reproduced the observed synergy and redundancy patterns, suggesting that such coding is a general feature of neural circuits.
• Stimulus features correlated with synergistic patterns were distinct and could not be predicted from individual cell responses, indicating unique encoding of stimulus structure.