The ability to experimentally perturb biological systems has traditionally been limited to static pre-programmed or operator-controlled protocols. In contrast, real-time control allows dynamic probing of biological systems with perturbations that are computed on-the-fly during experimentation. Real-time control applications for biological research are available; however, these systems are costly and often restrict the flexibility and customization of experimental protocols. The Real-Time eXperiment Interface (RTXI) is an open source software platform for achieving hard real-time data acquisition and closed-loop control in biological experiments while retaining the flexibility needed for experimental settings. RTXI has enabled users to implement complex custom closed-loop protocols in single cell, cell network, animal, and human electrophysiology studies. RTXI is also used as a free and open source, customizable electrophysiology platform in open-loop studies requiring online data acquisition, processing, and visualization. RTXI is easy to install, can be used with an extensive range of external experimentation and data acquisition hardware, and includes standard modules for implementing common electrophysiology protocols.

In order to study the ability of coupled neural oscillators to synchronize in the presence of intrinsic as opposed to synaptic noise, we constructed hybrid circuits consisting of one biological and one computational model neuron with reciprocal synaptic inhibition using the dynamic clamp. Uncoupled, both neurons fired periodic trains of action potentials. Most coupled circuits exhibited qualitative changes between one-to-one phase-locking with fairly constant phasic relationships and phase slipping with a constant progression in the phasic relationships across cycles. The phase resetting curve (PRC) and intrinsic periods were measured for both neurons, and used to construct a map of the firing intervals for both the coupled and externally forced (PRC measurement) conditions. For the coupled network, a stable fixed point of the map predicted phase locking, and its absence produced phase slipping. Repetitive application of the map was used to calibrate different noise models to simultaneously fit the noise level in the measurement of the PRC and the dynamics of the hybrid circuit experiments. Only a noise model that added history-dependent variability to the intrinsic period could fit both data sets with the same parameter values, as well as capture bifurcations in the fixed points of the map that cause switching between slipping and locking. We conclude that the biological neurons in our study have slowly-fluctuating stochastic dynamics that confer history dependence on the period. Theoretical results to date on the behavior of ensembles of noisy biological oscillators may require re-evaluation to account for transitions induced by slow noise dynamics.

Adaptively optimizing experiments has the potential to significantly reduce the number of trials needed to build parametric statistical models of neural systems. However, application of adaptive methods to neurophysiology has been limited by severe computational challenges. Since most neurons are high-dimensional systems, optimizing neurophysiology experiments requires computing high-dimensional integrations and optimizations in real time. Here we present a fast algorithm for choosing the most informative stimulus by maximizing the mutual information between the data and the unknown parameters of a generalized linear model (GLM) that we want to fit to the neuron's activity. We rely on important log concavity and asymptotic normality properties of the posterior to facilitate the required computations. Our algorithm requires only low-rank matrix manipulations and a two-dimensional search to choose the optimal stimulus. The average running time of these operations scales quadratically with the dimensionality of the GLM, making real-time adaptive experimental design feasible even for high-dimensional stimulus and parameter spaces. For example, we require roughly 10 milliseconds on a desktop computer to optimize a 100-dimensional stimulus. Despite using some approximations to make the algorithm efficient, our algorithm asymptotically decreases the uncertainty about the model parameters at a rate equal to the maximum rate predicted by an asymptotic analysis. Simulation results show that picking stimuli by maximizing the mutual information can speed up convergence to the optimal values of the parameters by an order of magnitude compared to using random (nonadaptive) stimuli. Finally, applying our design procedure to real neurophysiology experiments requires addressing the nonstationarities that we would expect to see in neural responses; our algorithm can efficiently handle both fast adaptation due to spike history effects and slow, nonsystematic drifts in a neuron's activity.

Sequential optimal design methods hold great promise for improving the efficiency of neurophysiology experiments. However, previous methods for optimal experimental design have incorporated only weak prior information about the underlying neural system (e.g., the sparseness or smoothness of the receptive field). Here we describe how to use stronger prior information, in the form of parametric models of the receptive field, in order to construct optimal stimuli and further improve the efficiency of our experiments. For example, if we believe that the receptive field is well-approximated by a Gabor function, then our method constructs stimuli that optimally constrain the Gabor parameters (orientation, spatial frequency, etc.) using as few experimental trials as possible. More generally, we may believe a priori that the receptive field lies near a known sub-manifold of the full parameter space; in this case, our method chooses stimuli in order to reduce the uncertainty along the tangent space of this sub-manifold as rapidly as possible. Applications to simulated and real data indicate that these methods may in many cases improve the experimental efficiency.

This work details CMOS, bio-inspired circuits that are used to link an artificial neuron and a living neuron and between two living neurons. We read an intracellular signal from a living neuron, use an integrate-and-fire neuron as a simple processing element to detect the spikes, and an artificial synapse to send outputs to a living neuron. The key structure is an electronic synapse which is based around a floating-gate pFET. The charge on the floating-gate is analogous to the synaptic weight and can be modified. This modification can be viewed as similar to LTP and LTD. Since programming technology has greatly improved, these weights can be set quickly and accurately

Adaptively optimizing experiments can significantly reduce the number of trials needed to characterize neural responses using parametric statistical models. However, the potential for these methods has been limited to date by severe computational challenges: choosing the stimulus which will provide the most information about the (typically high-dimensional) model parameters requires evaluating a high-dimensional integration and optimization in near-real time. Here we present a fast algorithm for choosing the optimal (most informative) stimulus based on a Fisher approximation of the Shannon information and specialized numerical linear algebra techniques. This algorithm requires only low-rank matrix manipulations and a one-dimensional linesearch to choose the stimulus and is therefore efficient even for high-dimensional stimulus and parameter spaces; for example, we require just 15 milliseconds on a desktop computer to optimize a 100-dimensional stimulus. Our algorithm therefore makes real-time adaptive experimental design feasible. Simulation results show that model parameters can be estimated much more efficiently using these adaptive techniques than by using random (nonadaptive) stimuli. Finally, we generalize the algorithm to efficiently handle both fast adaptation due to spike-history effects and slow, non-systematic drifts in the model parameters. Maximizing the efficiency of data collection is important in any experimental setting. In neurophysiology experiments, minimizing the number of trials needed to characterize a neural system is essential for maintaining the viability of a preparation and ensuring robust results. As a result, various approaches have been developed to optimize neurophysiology experiments online in order to choose the "best" stimuli given prior knowledge of the system and the observed history of the cell's responses. The "best" stimulus can be defined a number of different ways depending on the experimental objectives. One reasonable choice, if we are interested in finding a neuron's "preferred stimulus," is the stimulus which maximizes the firing rate of the neuron [1, 2, 3, 4]. Alternatively, when investigating the coding properties of sensory cells it makes sense to define the optimal stimulus in terms of the mutual information between the stimulus and response [5]. Here we take a system identification approach: we define the optimal stimulus as the one which tells us the most about how a neural system responds to its inputs [6, 7]. We consider neural systems in http://www.prism.gatech.edu/gtg120z http://www.stat.columbia.edu/liam which the probability p(rt |{xt , xt-1 , ..., xt-tk }, {rt-1 , . . . , rt-ta }) of the neural response rt given the current and past stimuli {xt , xt-1 , ..., xt-tk }, and the observed recent history of the neuron's activity, {rt-1 , . . . , rt-ta }, can be described by a model p(rt |{xt }, {rt-1 }, ), specified by a finite vector of parameters . Since we estimate these parameters from experimental trials, we want to choose our stimuli so as to minimize the number of trials needed to robustly estimate . Two inconvenient facts make it difficult to realize this goal in a computationally efficient manner: 1) model complexity -- we typically need a large number of parameters to accurately model a system's response p(rt |{xt }, {rt-1 }, ); and 2) stimulus complexity -- we are typically interested in neural responses to stimuli xt which are themselves very high-dimensional (e.g., spatiotemporal movies if we are dealing with visual neurons). In particular, it is computationally challenging to 1) update our a posteriori beliefs about the model parameters p(|{rt }, {xt }) given new stimulus-response data, and 2) find the optimal stimulus quickly enough to be useful in an online experimental context. In this work we present methods for solving these problems using generalized linear models (GLM) for the input-output relationship p(rt |{xt }, {rt-1 }, ) and certain Gaussian approximations of the posterior distribution of the model parameters. Our emphasis is on finding solutions which scale well in high dimensions. We solve problem (1) by using efficient rank-one update methods to update the Gaussian approximation to the posterior, and problem (2) by a reduction to a highly tractable onedimensional optimization problem. Simulation results show that the resulting algorithm produces a set of stimulus-response pairs which is much more informative than the set produced by random sampling. Moreover, the algorithm is efficient enough that it could feasibly run in real-time. Neural systems are highly adaptive and more generally nonstatic. A robust approach to optimal experimental design must be able to cope with changes in . We emphasize that the model framework analyzed here can account for three key types of changes: stimulus adaptation, spike rate adaptation, and random non-systematic changes. Adaptation which is completely stimulus dependent can be accounted for by including enough stimulus history terms in the model p(rt |{xt , ..., xt-tk }, {rt-1 , ..., rt-ta }). Spike-rate adaptation effects, and more generally spike history-dependent effects, are accounted for explicitly in the model (1) below. Finally, we consider slow, non-systematic changes which could potentially be due to changes in the health, arousal, or attentive state of the preparation. Methods We model a neuron as a point process whose conditional intensity function (instantaneous firing rate) is given as the output of a generalized linear model (GLM) [8, 9]. This model class has been discussed extensively elsewhere; briefly, this class is fairly natural from a physiological point of view [10], with close connections to biophysical models such as the integrate-and-fire cell [9], and has been applied in a wide variety of experimental settings [11, 12, 13, 14]. The model is summarized as: i ltk ( jta aj rt-j t = E (rt ) = f ki,t-l xi,t-l + 1) =1 =1 In the above summation the filter coefficients ki,t-l capture the dependence of the neuron's instantaneous firing rate t on the ith component of the vector stimulus at time t - l, xt-l ; the model therefore allows for spatiotemporal receptive fields. For convenience, we arrange all the stimulus coefficients in a vector, k , which allows for a uniform treatment of the spatial and temporal components of the receptive field. The coefficients aj model the dependence on the observed recent activity r at time t - j (these terms may reflect e.g. refractory effects, burstiness, firing-rate adaptation, etc., depending on the value of the vector a [9]). For convenience we denote the unknown parameter vector as = {k ; a}. The experimental objective is the estimation of the unknown filter coefficients, , given knowledge of the stimuli, xt , and the resulting responses rt . We chose the nonlinear stage of the GLM, the link function f (), to be the exponential function for simplicity. This choice ensures that the log likelihood of the observed data is a concave function of [9]. Representing and updating the posterior. As emphasized above, our first key task is to efficiently update the posterior distribution of after t trials, p(t |xt , rt ), as new stimulus-response pairs are trial 0 info. max.

Bursting, a dynamical phenomenon whereby episodes of neural action potentials are punctuated by periodic episodes of inactivity, is ubiquitous in neural systems. Examples include components of the respiratory rhythm generating circuitry in the brain stem, spontaneous activity in the neonatal rat spinal cord, and developing neural networks in the retina of the immature ferret. Bursting can also manifest itself in single neurons. Bursting dynamics require one or more kinetic processes slower than the timescale of the action potentials. Such processes usually manifest themselves in intrinsic ion channel properties, such as slow voltage-dependent gating or calcium-dependent processes, or synaptic mechanisms, such as synaptic depression. In this note, we show rhythmic bursting in a simulated neural network where no such slow processes exist at the cellular or synaptic level. Rather, the existence of rhythmic bursting is critically dependent on the connectivity of the network and manifests itself only when connectivity is characterized as small world. The slow process underlying the timescale of bursting manifests itself as a progressive synchronization of the network within each burst.

We present a reduction of a Hodgkin-Huxley (HH)--style bursting model to a hybridized integrate-and-fire (IF) formalism based on a thorough bifurcation analysis of the neuron's dynamics. The model incorporates HH--style equations to evolve the subthreshold currents and includes IF mechanisms to characterize spike events and mediate interactions between the subthreshold and spiking currents. The hybrid IF model successfully reproduces the dynamic behavior and temporal characteristics of the full model over a wide range of activity, including bursting and tonic firing. Comparisons of timed computer simulations of the reduced model and the original model for both single neurons and moderately sized networks (n < or = 500) show that this model offers improvement in computational speed over the HH--style bursting model.