Characterizing neural coding performance for populations of sensory neurons: comparing a weighted spike distance metrics to other analytical methods.

The identity of sensory stimuli is encoded in the spatio-temporal patterns of responses of the encoding neural population. For stimuli to be discriminated reliably, differences in population responses must be accurately decoded by downstream networks. Several methods to compare patterns of responses have been used by neurophysiologists to characterize the accuracy of the sensory responses studied. Among the most widely used analyses, we note methods based on Euclidean distances or on spike metric distances. Methods based on artificial neural networks and machine learning that recognize and/or classify specific input patterns have also gained popularity. Here, we first compare these three strategies using datasets from three different model systems: the moth olfactory system, the electrosensory system of gymnotids, and leaky-integrate-and-fire (LIF) model responses. We show that the input-weighting procedure inherent to artificial neural networks allows the efficient extraction of information relevant to stimulus discrimination. To combine the convenience of methods such as spike metric distances but leverage the advantages of weighting the inputs, we propose a measure based on geometric distances where each dimension is weighted proportionally to how informative it is. We show that the result of this Weighted Euclidian Distance (WED) analysis performs as well or better than the artificial neural network we tested and outperforms the more traditional spike distance metrics. We applied information theoretic analysis to LIF responses and compared their encoding accuracy with the discrimination accuracy quantified through this WED analysis. We show a high degree of correlation between discrimination accuracy and information content, and that our weighting procedure allowed the efficient use of information present to perform the discrimination task. We argue that our proposed measure provides the flexibility and ease of use sought by neurophysiologists while providing a more powerful way to extract relevant information than more traditional methods.

Modeling cancelation of periodic inputs with burst-STDP and feedback.

Prediction and cancelation of redundant information is an important feature that many neural systems must display in order to efficiently code external signals. We develop an analytic framework for such cancelation in sensory neurons produced by a cerebellar-like structure in wave-type electric fish. Our biologically plausible mechanism is motivated by experimental evidence of cancelation of periodic input arising from the proximity of conspecifics as well as tail motion. This mechanism involves elements present in a wide range of systems: (1) stimulus-driven feedback to the neurons acting as detectors, (2) a large variety of temporal delays in the pathways transmitting such feedback, responsible for producing frequency channels, and (3) burst-induced long-term plasticity. The bursting arises from back-propagating action potentials. Bursting events drive the input frequency-dependent learning rule, which in turn affects the feedback input and thus the burst rate. We show how the mean firing rate and the rate of production of 2- and 4-spike bursts (the main learning events) can be estimated analytically for a leaky integrate-and-fire model driven by (slow) sinusoidal, back-propagating and feedback inputs as well as rectified filtered noise. The effect of bursts on the average synaptic strength is also derived. Our results shed light on why bursts rather than single spikes can drive learning in such networks "online", i.e. in the absence of a correlative discharge. Phase locked spiking in frequency specific channels together with a frequency-dependent STDP window size regulate burst probability and duration self-consistently to implement cancelation.

Efficient inhibition of bursts by bursts in the auditory system of crickets.

In crickets, auditory information about ultrasound is carried bilaterally to the brain by the AN2 neurons. The ON1 neuron provides contralateral inhibitory input to AN2, thereby enhancing bilateral contrast between the left and right AN2s, an important cue for sound localization. We examine how the structures of the spike trains of these neurons affect this inhibitory interaction. As previously shown for AN2, ON1 responds to salient peaks in stimulus amplitude with bursts of spikes. Spike bursts, but not isolated spikes, reliably signal the occurrence of specific features of the stimulus. ON1 and AN2 burst at similar times relative to the amplitude envelope of the stimulus, and bursts are more tightly time-locked to stimulus feature than the isolated spikes. As a consequence, spikes that, in the absence of contralateral inhibition, would occur within AN2 bursts are more likely to be preceded by spikes in ON1 (mainly also in bursts) than are isolated AN2 spikes. This leads to a large decrease in the burst rate of the inhibited AN2. We conclude that the match in coding properties of ON1 and AN2 allows contralateral inhibition to be most efficient for those portions of the response that carry the behaviourally relevant information, i.e. for bursts.

Differential temporal coding of rhythmically diverse acoustic signals by a single interneuron.

The omega neuron 1 (ON1) of the cricket Teleogryllus oceanicus responds to conspecific signals (4.5 kHz) and to the ultrasonic echolocation sounds used by hunting, insectivorous bats. These signals differ in temporal structure as well as in carrier frequency. We show that ON1's temporal coding properties vary with carrier frequency, allowing it to encode both of these behaviorally important signals. Information-transfer functions show that coding of 4.5 kHz is limited to the range of amplitude-modulation components that occur in cricket songs (<32 Hz), whereas coding of 30-kHz stimuli extends to the higher pulse rates that occur in bat sounds ( approximately 100 Hz). Nonlinear coding contributes to the information content of ON1's spike train, particularly for 30-kHz stimuli with high intensities and large modulation depths. Phase locking to sinusoidal amplitude envelopes also extends to higher AM frequencies for ultrasound stimuli. ON1s frequency-specific behavior cannot be ascribed to differences in the shapes of information-transfer functions of low- and high-frequency-tuned receptor neurons, both of which are tuned more broadly to AM frequencies than ON1. Coding properties are nearly unaffected by contralateral deafferentation. ON1's role in auditory processing is to increase binaural contrast through contralateral inhibition. We hypothesize that its frequency-specific temporal coding properties optimize binaural contrast for sounds with both the spectral and temporal features of behaviorally relevant signals.