Function of identified interneurons in the leech elucidated using neural networks trained by back-propagation

Prediction of the Soil Permeability Coefficient of Reservoirs Using a Deep Neural Network Based on a Dendrite Concept

Changes in the pore water pressure of soil are essential factors that affect the movement of structures during and after construction in terms of stability and safety. Soil permeability represents the quantity of water transferred using pore water pressure. However, these changes cannot be easily identified and require considerable time and money. This study predicted and evaluated the soil permeability coefficient using a multiple regression (MR) model, adaptive network-based fuzzy inference system (ANFIS), general deep neural network (DNN) model, and DNN using the dendrite concept (DNN−T, which was proposed in this study). The void ratio, unit weight, and particle size were obtained from 164 undisturbed samples collected from the embankments of reservoirs in South Korea as input variables for the aforementioned models. The data used in this study included seven input variables, and the ratios of the training data to the validation data were randomly extracted, such as 6:4, 7:3, and 8:2, and were used. The analysis results for each model showed a median correlation of r = 0.6 or less and a low model efficiency of Nash–Sutcliffe efficiency (NSE) = 0.35 or less as a result of predicting MR and ANFIS. The DNN and DNN−T both have good performance, with a strong correlation of r = 0.75 or higher. Evidently, the DNN−T performance in terms of r, NSE, and root mean square error (RMSE) improved more than that of the DNN. However, the difference between the mean absolute percent error (MAPE) of DNN−T and the DNN was that the error of the DNN was small (11%). Regarding the ratio of the training data to the verification data, 7:3 and 8:2 showed better results compared to 6:4 for indicators, such as r, NSE, RMSE, and MAPE. We assumed that this phenomenon was caused by the DNN−T thinking layer. This study shows that DNN−T, which changes the structure of the DNN, is an alternative for estimating the soil permeability coefficient in the safety inspection of construction sites and is an excellent methodology that can save time and budget.

A tale of two leeches: Toward the understanding of the evolution and development of behavioral neural circuits

In the animal kingdom, behavioral traits encompass a broad spectrum of biological phenotypes that have critical roles in adaptive evolution, but an EvoDevo approach has not been broadly used to study behavior evolution. Here, we propose that, by integrating two leech model systems, each of which has already attained some success in its respective field, it is possible to take on behavioral traits with an EvoDevo approach. We first identify the developmental changes that may theoretically lead to behavioral evolution and explain why an EvoDevo study of behavior is challenging. Next, we discuss the pros and cons of the two leech model species, Hirudo, a classic model for invertebrate neurobiology, and Helobdella, an emerging model for clitellate developmental biology, as models for behavioral EvoDevo research. Given the limitations of each leech system, neither is particularly strong for behavioral EvoDevo. However, the two leech systems are complementary in their technical accessibilities, and they do exhibit some behavioral similarities and differences. By studying them in parallel and together with additional leech species such as Haementeria, it is possible to explore the different levels of behavioral development and evolution.

Zebrafish Neuroscience: Using Artificial Neural Networks to Help Understand Brains

Brains are notoriously hard to understand, and neuroscientists need all the tools they can get their hands on to have a realistic shot at it. Advances in machine learning are proving instrumental, illustrated by their recent use to shed light on navigational strategies implemented by zebrafish brains.

Artificial neural networks accurately predict intra-abdominal infection in moderately severe and severe acute pancreatitis

OBJECTIVE

The aim of this study was to evaluate the efficacy of artificial neural networks (ANN) in predicting intra-abdominal infection in moderately severe (MASP) and severe acute pancreatitis (SAP) compared with that of a logistic regression model (LRM).

METHODS

Patients suffering from MSAP or SAP from July 2014 to June 2017 in three affiliated hospitals of the Army Medical University in Chongqing, China, were enrolled in this study. A univariate analysis was used to determine the different parameters between patients with and without intra-abdominal infection. Subsequently, these parameters were used to build LRM and ANN.

RESULTS

Altogether 263 patients with MSAP or SAP were enrolled in this retrospective study. A total of 16 parameters that differed between patients with and without intra-abdominal infection were used to construct both models. The sensitivity of ANN and LRM was 80.99% (95% confidence interval [CI] 72.63-87.33) and 70.25% (95% CI 61.15-78.04), respectively (P > 0.05), whereas the specificity was 89.44% (95% CI 82.89-93.77) and 77.46% (95% CI 69.54-83.87), respectively (P < 0.05). ANN predicted the risk of intra-abdominal infection better than LRM (area under the receiver operating characteristic curve: 0.923 [0.883-0.952] vs 0.802 [0.749-0.849], P < 0.001).

CONCLUSIONS

ANN accurately predicted intra-abdominal infection in MSAP and SAP and is an ideal tool for predicting intra-abdominal infection in such patients. Coagulation parameters played an important role in such prediction.

Convergent Temperature Representations in Artificial and Biological Neural Networks

Discoveries in biological neural networks (BNNs) shaped artificial neural networks (ANNs) and computational parallels between ANNs and BNNs have recently been discovered. However, it is unclear to what extent discoveries in ANNs can give insight into BNN function. Here, we designed and trained an ANN to perform heat gradient navigation and found striking similarities in computation and heat representation to a known zebrafish BNN. This included shared ON- and OFF-type representations of absolute temperature and rates of change. Importantly, ANN function critically relied on zebrafish-like units. We furthermore used the accessibility of the ANN to discover a new temperature-responsive cell type in the zebrafish cerebellum. Finally, constraining the ANN by the C. elegans motor repertoire retuned sensory representations indicating that our approach generalizes. Together, these results emphasize convergence of ANNs and BNNs on stereotypical representations and that ANNs form a powerful tool to understand their biological counterparts.

Effects of Touch Location and Intensity on Interneurons of the Leech Local Bend Network

Touch triggers highly precise behavioural responses in the leech. The underlying network of this so-called local bend reflex consists of three layers of individually characterised neurons. While the population of mechanosensory cells provide multiplexed information about the stimulus, not much is known about how interneurons process this information. Here, we analyse the responses of two local bend interneurons (cell 157 and 159) to a mechanical stimulation of the skin and show their response characteristics to naturalistic stimuli. Intracellular dye-fills combined with structural imaging revealed that these interneurons are synaptically coupled to all three types of mechanosensory cells (T, P, and N cells). Since tactile stimulation of the skin evokes spikes in one to two cells of each of the latter types, interneurons combine inputs from up to six mechanosensory cells. We find that properties of touch location and intensity can be estimated reliably and accurately based on the graded interneuron responses. Connections to several mechanosensory cell types and specific response characteristics of the interneuron types indicate specialised filter and integration properties within this small neuronal network, thus providing evidence for more complex signal processing than previously thought.

Recurrent inhibition in motor systems, a comparative analysis

The review proposes a comparison between recurrent inhibition in motor systems of vertebrates and the leech nervous system, where a detailed cellular and functional analysis has been accomplished. A comparative study shows that recurrent inhibition is a conserved property in motor systems of phylogenetically distant species. Recurrent inhibition has been extensively characterized in the spinal cord of mammals, where Renshaw cells receive excitatory synaptic inputs from motoneurons (MNs) and, in turn, exert an inhibitory effect on the MNs. In the leech, a recurrent inhibitory circuit has been described, centered around a pair of nonspiking (NS) neurons. NS are linked to every excitatory MN through rectifying electrical junctions. And, in addition, the MNs are linked to the NS neurons through hyperpolarizing chemical synapses. Functional analysis of this leech circuit showed that heteronymous MNs in the leech are electrically coupled and this coupling is modulated by the membrane potential of NS neurons. Like Renshaw cells, the membrane potential of NS neurons oscillates in phase with rhythmic motor patterns. Functional analysis performed in the leech shows that NS influences the activity of MNs in the course of crawling suggesting that the recurrent inhibitory circuit modulates the motor performance.

Inducible and titratable silencing of Caenorhabditis elegans neurons in vivo with histamine-gated chloride channels

Recent progress in neuroscience has been facilitated by tools for neuronal activation and inactivation that are orthogonal to endogenous signaling systems. We describe here a chemical-genetic approach for inducible silencing of Caenorhabditis elegans neurons in intact animals, using the histamine-gated chloride channel HisCl1 from Drosophila and exogenous histamine. Administering histamine to freely moving C. elegans that express HisCl1 transgenes in neurons leads to rapid and potent inhibition of neural activity within minutes, as assessed by behavior, functional calcium imaging, and electrophysiology of neurons expressing HisCl1. C. elegans does not use histamine as an endogenous neurotransmitter, and exogenous histamine has little apparent effect on wild-type C. elegans behavior. HisCl1-histamine silencing of sensory neurons, interneurons, and motor neurons leads to behavioral effects matching their known functions. In addition, the HisCl1-histamine system can be used to titrate the level of neural activity, revealing quantitative relationships between neural activity and behavioral output. We use these methods to dissect escape circuits, define interneurons that regulate locomotion speed (AVA, AIB) and escape-related omega turns (AIB), and demonstrate graded control of reversal length by AVA interneurons and DA/VA motor neurons. The histamine-HisCl1 system is effective, robust, compatible with standard behavioral assays, and easily combined with optogenetic tools, properties that should make it a useful addition to C. elegans neurotechnology.

Developmentally regulated multisensory integration for prey localization in the medicinal leech

Medicinal leeches, like many aquatic animals, use water disturbances to localize their prey, so they need to be able to determine if a wave disturbance is created by prey or by another source. Many aquatic predators perform this separation by responding only to those wave frequencies representing their prey. As leeches' prey preference changes over the course of their development, we examined their responses at three different life stages. We found that juveniles more readily localize wave sources of lower frequencies (2 Hz) than their adult counterparts (8-12 Hz), and that adolescents exhibited elements of both juvenile and adult behavior, readily localizing sources of both frequencies. Leeches are known to be able to localize the source of waves through the use of either mechanical or visual information. We separately characterized their ability to localize various frequencies of stimuli using unimodal cues. Within a single modality, the frequency-response curves of adults and juveniles were virtually indistinguishable. However, the differences between the responses for each modality (visual and mechanosensory) were striking. The optimal visual stimulus had a much lower frequency (2 Hz) than the optimal mechanical stimulus (12 Hz). These frequencies matched, respectively, the juvenile and the adult preferred frequency for multimodally sensed waves. This suggests that, in the multimodal condition, adult behavior is driven more by mechanosensory information and juvenile behavior more by visual. Indeed, when stimuli of the two modalities were placed in conflict with one another, adult leeches, unlike juveniles, were attracted to the mechanical stimulus much more strongly than to the visual stimulus.

Physiology and morphology of shared and specialized spinal interneurons for locomotion and scratching

Distinct types of rhythmic movements that use the same muscles are typically generated largely by shared multifunctional neurons in invertebrates, but less is known for vertebrates. Evidence suggests that locomotion and scratching are produced partly by shared spinal cord interneuronal circuity, although direct evidence with intracellular recording has been lacking. Here, spinal interneurons were recorded intracellularly during fictive swimming and fictive scratching in vivo and filled with Neurobiotin. Some interneurons that were rhythmically activated during both swimming and scratching had axon terminal arborizations in the ventral horn of the hindlimb enlargement, indicating their likely contribution to hindlimb motor outputs during both behaviors. We previously described a morphological group of spinal interneurons ("transverse interneurons" or T neurons) that were rhythmically activated during all forms of fictive scratching at higher peak firing rates and with larger membrane potential oscillations than scratch-activated spinal interneurons with different dendritic orientations. The current study demonstrates that T neurons are activated during both swimming and scratching and thus are components of the shared circuitry. Many spinal interneurons activated during fictive scratching are also activated during fictive swimming (scratch/swim neurons), but others are suppressed during swimming (scratch-specialized neurons). The current study demonstrates that some scratch-specialized neurons receive strong and long-lasting hyperpolarizing inhibition during fictive swimming and are also morphologically distinct from T neurons. Thus this study indicates that locomotion and scratching are produced by a combination of shared and dedicated interneurons whose physiological and morphological properties are beginning to be revealed.

Neuronal control of leech behavior

The medicinal leech has served as an important experimental preparation for neuroscience research since the late 19th century. Initial anatomical and developmental studies dating back more than 100 years ago were followed by behavioral and electrophysiological investigations in the first half of the 20th century. More recently, intense studies of the neuronal mechanisms underlying leech movements have resulted in detailed descriptions of six behaviors described in this review; namely, heartbeat, local bending, shortening, swimming, crawling, and feeding. Neuroethological studies in leeches are particularly tractable because the CNS is distributed and metameric, with only 400 identifiable, mostly paired neurons in segmental ganglia. An interesting, yet limited, set of discrete movements allows students of leech behavior not only to describe the underlying neuronal circuits, but also interactions among circuits and behaviors. This review provides descriptions of six behaviors including their origins within neuronal circuits, their modification by feedback loops and neuromodulators, and interactions between circuits underlying with these behaviors.

Computational subunits of visual cortical neurons revealed by artificial neural networks

A crucial step toward understanding visual processing is to obtain a comprehensive description of the relationship between visual stimuli and neuronal responses. Many neurons in the visual cortex exhibit nonlinear responses, making it difficult to characterize their stimulus-response relationships. Here, we recorded the responses of primary visual cortical neurons of the cat to spatiotemporal random-bar stimuli and trained artificial neural networks to predict the response of each neuron. The random initial connections in the networks consistently converged to regular patterns. Analyses of these connection patterns showed that the response of each complex cell to the random-bar stimuli could be well approximated by the sum of a small number of subunits resembling simple cells. The direction selectivity of each complex cell measured with drifting gratings was also well predicted by the combination of these subunits, indicating the generality of the model. These results are consistent with a simple functional model for complex cells and demonstrate the usefulness of the neural network method for revealing the stimulus-response transformations of nonlinear neurons.

Artificial neural networks estimate the contribution of taste neurons to coding

Taste qualities are believed to be coded in the activity of populations of taste neurons. However, it is not clear whether all neurons are equally responsible for coding. To clarify the point the relative contribution of each taste neuron to coding was assessed by constructing simple three-layer neural networks with input neurons that represent cortical taste neurons of the rat. The networks were trained by the back-propagation learning algorithm to classify the neural response patterns to the basic taste stimuli (sucrose, HCl, quinine-hydrochloride, and NaCl). The networks had four output neurons representing the basic taste qualities, the values of which provide a measure for similarity of test stimuli to the basic taste stimuli. We estimated relative contributions of input neurons to the taste discrimination of the network by examining their significance S(j), which is defined as the sum of the absolute values of the connection weights from the jth input neuron to the hidden layer. When the input neurons with a smaller S(j) (e.g., 15 out of 39 input units) were "pruned" from the trained network, the ability of the network to discriminate the basic taste qualities was not greatly affected. On the other hand, the taste discrimination of the network progressively deteriorated much more rapidly with pruning of input neurons with a larger S(j). These results suggest that cortical taste neurons differentially contribute to the coding of taste qualities. Input neurons with a larger S(j) tended to be with a larger variation of neural discharge rates to the basic taste stimuli. The variation of neural discharges may be important in the coding of taste qualities.

A neuronal network for computing population vectors in the leech

The correlation of neuronal activity with sensory input and behavioural output has revealed that information is often encoded in the activity of many neurons across a population, that is, a neural population code is used. The possible algorithms that downstream networks use to read out this population code have been studied by manipulating the activity of a few neurons in a population. We have used this approach to study population coding in a small network underlying the leech local bend, a body bend directed away from a touch stimulus. Because of the small size of this network we are able to monitor and manipulate the complete set of sensory inputs to the network. We show here that the population vector formed by the spike counts of the active mechanosensory neurons is well correlated with bend direction. A model based on the known connectivity of the identified neurons in the local bend network can account for our experimental results, and is suitable for reading out the neural population vector. Thus, for the first time to our knowledge, it is possible to link a proposed algorithm for neural population coding with synaptic and network mechanisms in an experimental system.

Population coding and behavioral choice

Many individual behavioral acts are produced by the combined activity of large populations of broadly tuned neurons, and the neuronal populations for different behaviors can overlap. Recent experiments monitoring and manipulating neuronal activity during behavioral decisions have begun to shed light on the mechanisms that enable overlapping populations of neurons to generate choices between categorically distinct behaviors.

What we have learned from the study of learning in the leech

The use of invertebrate preparations has contributed greatly to our understanding of the neural basis of learning. The leech is especially useful for studying behavioral changes and their underlying neuronal mechanisms. Learning in the leech is essentially identical to that found in other animals, both vertebrate and invertebrate. Using anatomical and physiological techniques on leeches as they learn, we have begun to characterize the properties of individual neurons and neuronal networks that play a role in learning. We have been able to show two neuronal mechanisms that have not been previously associated with associative conditioning. The first has to do with the importance of contingency: one stimulus [the conditional stimulus (CS)] becomes associated with a second stimulus [the unconditional stimulus, (US)] in proportion to the ability of the CS to predict the US. We have found that important properties for encoding predictability, such as circuit reconfiguration, may lie in the US pathway. The firing of the serotonergic Retzius cells is taken as the US; consistent CS prediction of a US prevents "dropout" of a critical component of one US pathway. Throughout training, predicted USs continue to elicit a barrage of action potentials in these cells. Recurring unpredicted USs degrade both the learning and the response of the Retzius cell to the US. A second insight is that at least two US pathways contribute to learning, the Retzius cell pathway and the nociceptive (N) cell pathway. This second pathway persists after the elimination of the Retzius cell pathway. The observation of multiple US pathways raises a host of issues concerning CS-US convergence and the functional significance of distinct US pathways, and our results are discussed in terms of implications to current models of learning.

Using reflexive behaviors of the medicinal leech to study information processing

The interneuronal network that produces local bending in the leech is distributed, in the sense that most of the interneurons involved are activated in all forms of local bending, even those in which their outputs would produce inappropriate movements. Such networks have been found to control a number of different behaviors in a variety of animals. This article reviews three issues: the physiological and modeling observations that led to the conclusion that local bending in leeches is controlled by a distributed system; what distributed processing means for this and other behaviors; and why the leech interneuronal network may have evolved to be distributed in the first place.

Computer-simulated neural networks: an appropriate model for motor development?

The idea of an artificial neural network is introduced in a historical context, and the essential aspect of it, viz., the modifiable synapse, is compared to the aspect of plasticity in the natural nervous system. Based on such an artificial neural network, a model is presented for the way in which (the path along which) the connectivity in the spinal cord is modified during the period that a newborn 'learns' to control the movement of his forearm. In this way an automatic calibration of the receptors and the antagonists' recruitment of motor units is represented. The learning process is described in non-mathematical terminology. The model is then shown to be able after learning to reach target angles outside the training set of angles, and to be able to relearn when an important receptor has been made inoperative. In this way it is shown that the model is able to generalize, and that it is robust against at least some damage.

A lower bound on the detectability of nonassociative learning in the local bending reflex of the medicinal leech

Studies of neural mechanisms of learning and memory have focused on large changes at identified synapses. However, memory in distributed processing reflexes could involve widely distributed engrams characterized by small changes at every synapse in the network. To investigate this possibility, we used a neural network optimization algorithm to construct distributed engrams for nonassociative conditioning in a model of the local bending reflex of the medicinal leech (Hirudo medicinalis). The model comprised 4 sensory neurons, 10 to 40 interneurons, 8 motor neurons, and up to 480 connections. Synaptic connections in the model were first optimized to reproduce the amplitude and time course of motor neuron synaptic potentials recorded during local bending. This network, which represented the naive state before conditioning, was then reoptimized to the habituated or sensitized state. Following reoptimization, the memory for nonassociative learning was encoded by small changes dispersed across the entire network, and each change made only a small contribution to the learning. Moreover, because the changes were small, resolution of a few tenths of a millivolt, or 3-5% of an average synaptic potential, would be needed to account for half of the nonassociative learning. These results show how difficult distributed engrams can be to detect and provide a likely lower bound on the detectability of nonassociative learning in this and related networks.

Synaptic interactions among neurons that coordinate swimmeret and abdominal movements in the crayfish

1. Many interneurons in the crayfish (Procambarus clarkii) abdominal nervous system influence two behaviors, abdominal positioning and swimmeret movements. Such neurons are referred to as dual output cells. Other neurons which influence either one behavior or the other are single output cells. 2. Extensive synaptic interactions were observed between both dual and single output neurons involved in the control of abdominal positioning and swimmeret movements. Over 60% of all neuron pairs examined displayed interactions. Pairs of agonist neurons displayed excitatory interactions, while pairs of antagonists had inhibitory interactions. This pattern of interaction was observed in about 75% of interactive neuron pairs whether abdominal positioning or swimmeret outputs were considered. 3. Evidence for both serial and parallel connectivity, as well as, reciprocal or looping connections was observed. Looping connections can be found both between the abdominal positioning and swimmeret systems and within each system. 4. Most (28/34) single output neurons were not presynaptic to dual output neurons. No single output neurons were found to excite dual output neurons to spiking, although inhibitory interactions and weak excitations were observed. 5. Abdominal positioning inhibitors displayed properties consistent with a role in mediating some of the coordination between the swimmeret and abdominal positioning systems. 6. None of the dual output neurons examined influenced the swimmeret motoneurons directly.

Two forms of sensitization of the local bending reflex of the medicinal leech

Sensitization of the local bending reflex of the medicinal leech Hirudo medicinalis was studied in a semi-intact preparation in which behavioral and electrophysiological recordings were made simultaneously. 1. Sensitization of local bending could be produced in two ways: by repeated stimulation of the mechanoreceptor sensitive to pressure (the P cell), and by stimulation of the mechanoreceptor sensitive to noxious stimuli (the N cell). 2. Both forms of sensitization produced a central neuronal change, measured as an increase in the number of stimulus-evoked action potentials in cell 3 (an excitor of dorsal longitudinal muscles). 3. Intracellular stimulation of serotonin-containing neurons 21 and 61 mimicked the sensitizing stimuli, but stimulation of the Retzius cell, which also contains serotonin, did not. 4. Stimulation of the Leydig cell, which releases octopamine, decreased the strength of local bending.