Timezone: »

 
Workshop
Workshop on Worm's Neural Information Processing (WNIP)
Ramin Hasani · Manuel Zimmer · Stephen Larson · Tomas Kazmar · Radu Grosu

Fri Dec 08 08:00 AM -- 06:30 PM (PST) @ S5
Event URL: https://sites.google.com/site/wwnip2017/ »

A fundamental Challenge in neuroscience is to understand the elemental computations and algorithms by which brains perform information processing. This is of great significance to biologists, as well as, to engineers and computer scientists, who aim at developing energy efficient and intelligent solutions for the next generation of computers and autonomous devices. The benefits of collaborations between these fields are reciprocal, as brain-inspired computational algorithms and devices not only advance engineering, but also assist neuroscientists by conforming their models and making novel predictions. A large impediment toward such an efficient interaction is still the complexity of brains. We thus propose that the study of small model organisms should pioneer these efforts.

The nematode worm, C. elegans, provides a ready experimental system for reverse-engineering the nervous system, being one of the best studied animals in the life sciences. The neural connectome of C. elegans has been known for 30 years, providing the structural basis for building models of its neural information processing. Despite its small size, C. elegans exhibits complex behaviors, such as, locating food, mating partners and navigating its environment by integrating a plethora of environmental cues. Over the past years, the field has made an enormous progress in understanding some of the neural circuits that control sensory processing, decision making and locomotion. In laboratory, the crawling behavior of worms occurs mainly in 2D. This enables the use of machine learning tools to obtain quantitative behavioral descriptions of unprecedented accuracy. Moreover, neuronal imaging techniques have been developed so that the activity of nearly all nerve cells in the brain can be recorded in real time. Leveraging on these advancements, the community wide C. elegans OpenWorm project will make a realistic in silico simulation of a nervous system and the behavior it produces possible, for the first time. 

The goal of this workshop is to gather researchers in neuroscience and machine learning together, to advance understanding of the neural information processing of the worm and to outline what challenges still lie ahead. We particularly aim to:
- Comprehensively, introduce the nervous system of C. elegans. We will discuss the state-of-the-art findings and potential future solutions for modeling its neurons and synapses, complete networks of neurons and the various behaviors of the worm,
- Identify main challenges and their solutions in behavioral and neural data extraction, such as imaging techniques, generation of time series data from calcium imaging records and high resolution behavioral data, as well as cell recognition, cell tracking and image segmentation,
- Explore machine learning techniques for interpretation of brain data, such as time series analysis, feature extraction methods, complex network analysis, complex nonlinear systems analysis, large-scale parameter optimization methods, and representation learning,
- Get inspirations from this well-understood brain to design novel network architectures, control algorithms and neural processing units.
We have invited leading neuroscientists, machine learning scientists and interdisciplinary experts, to address these main objectives of the workshop, in the form of Keynote talks and a panel discussion. We also invite submissions of 4-page papers for posters, spotlight presentations and contributed talks, and offer travel awards.

Topics of interests are: Deep learning applications in nervous system data analysis, neural circuits analysis, behavior modeling, novel computational approaches and algorithms for brain data interpretations, brain simulation platforms, optimization algorithms for nonlinear systems, applications of machine learning methods to brain data and cell biology, complex network analysis, cell modeling, cell recognition and tracking, dynamic modeling of neural circuits and genetic regulatory networks.

The workshop’s webpage: https://sites.google.com/site/wwnip2017/

Fri 9:00 a.m. - 9:15 a.m. [iCal]
Openning Remarks (Talk)
Ramin Hasani
Fri 9:15 a.m. - 9:45 a.m. [iCal]

While animal behavior is often quantified through discrete motifs, this is only an approximation to fundamentally continuous dynamics and ignores important variability within each motif. Here, we develop a behavioral phase space in which the instantaneous state is smoothly unfolded as a combination of postures and their short-time dynamics. We apply this approach to C. elegans and show that the dynamics lie on a 6D space, which is globally composed of three sets of cyclic trajectories that form the animal’s basic behavioral motifs: forward, backward and turning locomotion. In contrast to global stereotypy, variability is evident by the presence of locally-unstable dynamics for each set of cycles. Across the full phase space we show that the Lyapunov spectrum is symmetric with positive, chaotic exponents driving variability balanced by negative, dissipative exponents driving stereotypy. The symmetry of the spectrum holds for different environments and for human walking, suggesting a general condition of motor control. Finally, we use the reconstructed phase space to analyze the complexity of the dynamics along the worm’s body and find evidence for multiple, spatially-separate oscillators driving C. elegans locomotion.

Greg Stephens
Fri 9:45 a.m. - 10:15 a.m. [iCal]

Decision making is a central function of the brain and the focus of intensive study in neuroscience, psychology, and economics. Value-based decision making (e.g., ‘which fragrance do you prefer?’ not 'which smells more like roses?') guides significant, sometimes life-changing, choices yet its neuronal basis is poorly understood. Research into this question would be accelerated by the introduction of genetically tractable invertebrates with small nervous systems, like the Drosophila and C. elegans. We have recently shown that the nematode C. elegans makes value-based decisions. This was done using a formal economic method – the Generalized Axiom of Revealed Preference (GARP). The basis of the method is to establish that the subject's choices are internally consistent with respect to transitivity (A > B > C ⇒ A > C). In the wild, C. elegans feeds on a variety of bacteria and learns to prefer the more nutritious species. We tested worms on a set of decisions between a high quality species and a low quality species at a range of relative concentrations and found the worm’s choices to be 100% transitive, the necessary and sufficient condition for value-based decision making. Further, we found that the olfactory neuron AWC, known to be activated by the sudden absence of food, is required for intact food choice behavior. Surprisingly, however, we found that AWC is also activated by the switch from high quality food to low quality food, even when the two foods are at the same concentration. Thus, food value may be represented at the level of individual olfactory neurons. We are now investigating the neural mechanisms of choice transitivity. C. elegans selects food sources utilizing klinotaxis, a chemotaxis strategy during locomotion in which the worm’s head bends more deeply on the side of preferred food. The chemosensory neurons, interneurons, and motor neurons of a candidate circuit for klinotaxis have been identified. Extrapolating from our findings with respect to AWC, we have developed a model of the circuit in which distinct chemosensory neuron types encode food quality and quantity during particular phases of head bending. Activation of downstream interneurons in the model is the weighted sum of these inputs in accordance with phase information. The model proposes that the signs and strengths of synaptic weights in the biological circuit are adjusted to ensure that subjective value is a monotonic function of the relative quantity of high and low quality food, a property that guarantees transitivity under GARP. Work in progress tests the model using calcium imaging, optogenetic activation, and ablations of each neuron in the circuit.

Shawn Lockery
Fri 10:15 a.m. - 10:30 a.m. [iCal]
3 spotlight presentations (Spotlight Presentations)
Kelly Buchanan, Mathias Lechner, Kezhi Li
Fri 10:30 a.m. - 11:00 a.m. [iCal]
Poster Session 1 (Break / Poster Session)
Magdalena Fuchs, David Lung, Mathias Lechner, Kezhi Li, Andrew Gordus, Vivek Venkatachalam, Shivesh Chaudhary, Jan Hůla, David Rolnick, Scott Linderman, Gonzalo Mena, Liam Paninski, Netta Cohen
Fri 11:00 a.m. - 11:30 a.m. [iCal]

Populations of neurons in the brains of many different animals, ranging from invertebrates to primates, typically coordinate their activities to generate low dimensional and transient activity dynamics, an operational principle serving many neuronal functions like sensory coding, decision making and motor control. However, the mechanism that bind individual neurons to global population states are not yet known. Are population dynamics driven by a smaller number of pacemaker neurons or are they an emergent property of neuronal networks? What are the features in global network architecture that support coordinated network wide dynamics? In order to address these problems, we study neuronal population dynamics in C. elegans. We recently developed a calcium imaging approach to record the activity of nearly all neuron in the worm brain in real time and at single cell resolution. We show that brain activity of C. elegans is dominated by brain wide coordinated population dynamics involving a large fraction of interneurons and motor neurons. The activity patterns of these neuronal ensembles recur in an orderly and cyclical fashion. In subsequent experiments, we characterized these brain dynamics functionally and found that they represent action commands and their assembly into a typical action sequence of these animals: forward crawling – backward crawling – turning. Deciphering the mechanisms underlying neuronal population dynamics is key to understanding the principal computations performed by neuronal networks in the brains of animals, and perhaps will inspire the design of novel machine learning algorithms for robotic control. In this talk, I will discuss three of our approaches to uncover these mechanisms: First, using graph theory, we aim to identify the key features of neuronal network architecture that support functional dynamics. We found that rich club neurons, i.e. highly interconnected network hubs contribute most to brain dynamics. However, simple measures of synaptic connectivity (e.g. connection strength) failed to predict functional interactions between these neurons; unlike higher order network statistics that measure the similarity in synaptic input patterns. We next performed systematic perturbations by interrogation of rich club neurons via transgenic neuronal inhibition tools. Using whole brain imaging in combination with computational analysis methods we found that upon inhibition of critical hubs, leading to a disintegration of the network, most other individual neurons remain vigorously active, however the global coordination across neurons was abolished. Based on these results we hypothesize that neuronal population dynamics are an emergent property of neuronal networks. Finally, we aim to recapitulate C. elegans brain dynamics in silico. Here, we generate neuronal network simulations based on deterministic and stochastic biophysical models of neurons and synapses, at multiscale levels of abstraction. We then adopt a genetic algorithm for neuronal circuit parameter optimization, to find the best matches between simulations and measured calcium dynamics. This approach enables us to test our hypotheses, and to predict unknown properties of neural circuits important for brain dynamics.

Manuel Zimmer
Fri 11:30 a.m. - 12:00 p.m. [iCal]

Effective spatial navigation is essential for the survival of animals. Navigation, or the search for favorable conditions, is fundamentally an adaptive behavior that can depend on the changing environment, the animal's past history of success and failure and its internal state. C. elegans implements combinations of systematic and stochastic navigational strategies that are modulated by plasticity across a range of time scales. Here, we combine experiments and computational modeling to characterise adaptation in gustatory and nociceptive salt sensing neurons and construct a simulation framework in which animals can navigate a virtual environment. Our model, and simulations on a variety of smooth, rugged or complex landscapes, suggest that these different forms of sensory adaptation combine to dynamically modulate navigational strategies, giving rise to effective exploration and navigation of the environment. Inspired by this compact and elegant sensory circuit, we present a robotic simulation framework, capable of robustly searching for landmarks in a toy simulation environment.

Netta Cohen
Fri 12:00 p.m. - 12:15 p.m. [iCal]
Multi-neuronal imaging of C. elegans courtship and mating (Contributed Talk)
Vivek Venkatachalam
Fri 12:15 p.m. - 12:30 p.m. [iCal]

William R. Schafer1, Gang Yan2, 3, Petra E. Vértes4, Emma K. Towlson3, Yee Lian Chew1, Denise S. Walker1, & Albert-László Barabási3 1Division of Neurobiology, MRC Laboratory of Molecular Biology, Cambridge Biomedical Campus, Francis Crick Avenue, Cambridge CB2 0QH, UK. 2School of Physics Science and Engineering, Tongji University, Shanghai 200092, China. 3Center for Complex Network Research and Department of Physics, Northeastern University, Boston, Massachusetts 02115, USA. 4Department of Psychiatry, Behavioural and Clinical Neuroscience Institute, University of Cambridge, Cambridge CB2 0SZ, UK.

Large-scale efforts are underway to map the neuronal connectomes of many animals, from flies to humans. However, even for small connectomes, such as that of C. elegans, it has been difficult to relate the structure of neuronal wiring patterns to the function of neural circuits. Recent theoretical studies have suggested that control theory might provide a framework to understand structure-function relationships in complex biological networks, including neuronal connectomes. To test this hypothesis experimentally, we have used the complete neuronal connectome of C. elegans to identify neurons predicted to affect the controllability of the body muscles and assess the effect of ablating these neurons on locomotor behavior. We identified 12 neural classes whose removal from the connectome reduced the structural controllability of the body neuromusculature, one of which was the uncharacterized PDB motorneuron. Consistent with the control theory prediction, ablation of PDB had a specific effect on locomotion, altering the dorsoventral polarity of large turns. Control analysis also predicted that three members of the DD motorneuron class (DD4, DD5 and DD6) are individually required for body muscle controllability, while more anterior DDs (DD1, DD2 and DD3) are not. Indeed, we found that ablation of DD4 or DD5, but not DD2 or DD3, led to abnormalities in posterior body movements, again consistent with control theory predictions. We are currently using the control framework to probe other parts of the C. elegans connectome, and are developing more sophisticated approaches behavioral analysis in order to more precisely relate ablation phenotypes to specific muscle groups. We anticipate that the control framework validated by this work may have application in the analysis of larger neuronal connectomes and other complex networks.

William Schafer
Fri 12:30 p.m. - 2:00 p.m. [iCal]
Lunch Break (Break)
Fri 2:00 p.m. - 2:30 p.m. [iCal]

The membrane potential of a biological neuron is considered to be one of the most importantproperties to understand its dynamic state. While the action potential or discrete “spike” featureof mammalian neurons has been emphasized as an information bearing signal, biologicalevidence exists that even without action potentials, neurons process information and give rise todifferent behavioral states. Nowhere is this more evident than in the nematode worm C.elegans , where its entire nervous system of 302 neurons, despite a lack of action potentials,organizes complex behaviors such as mating, predator avoidance, location of food sources, andmany others. For thirty years, the C. elegans nervous system has remained the only adult animal that has hadits nervous system connectivity mapped at the level of individual synapses and gap junctions.As part of the international open science collaboration known as OpenWorm, we have built a1simulation framework, known as c302, that enables us to assemble the known connectivity andother biological data of the C. elegans nervous system into a Hodgkin­Huxley­based simulationthat can be run in the NEURON simulation engine. Using a physical simulation of the C. elegans body, known as Sibernetic, we have injected simple sinusoidal activation patterns of the muscle cells of the C. elegans and produced simplecrawling and swimming behavior. With the goal of producing the same simple sinusoids in themuscle cells, we have used c302 to select a subnetwork from the full C. elegans nervoussystem and used machine learning techniques to fit dynamic parameters that are underspecifiedby the data. Our preliminary results still leave many important biological features out, butinitially demonstrate that it is possible to make motor neurons produce sinusoidal activitypatterns in the muscles as used in the physical simulation.

In this talk I will discuss these initial results and discuss future directions for a betterunderstanding of the information processing underlying the C. elegans’ nervous system.

Stephen Larson
Fri 2:30 p.m. - 3:00 p.m. [iCal]
Parking with a worm's brain (Talk)
Radu Grosu
Fri 3:00 p.m. - 3:30 p.m. [iCal]
Break / Poster Session 2 (Break / Poster Session)
Fri 3:30 p.m. - 4:00 p.m. [iCal]

One of the grand scientific challenges of this century is to understand how behavior is grounded in the interaction between an organism’s brain, its body, and its environment. Although a lot of attention and resources are focused on understanding the human brain, I will argue that the study of simpler organisms are an ideal place to begin to address this challenge. I will introduce the nematode worm Caernohabditis elegans, with just 302 neurons, the only fully-reconstructed connectome at the cellular level, and a rich behavioral repertoire that we are still discovering. I will describe a computational approach to address such grand challenge. I will lay out some of the advantages of expressing our understanding in equations and computational models rather than just words. I will describe our unique methodology for exploring the unknown biological parameters of the model through the use of evolutionary algorithms. We train the neural networks on what they should do, with little or no instructions on how to do it. The effort is then to analyze and understand the evolved solutions as a way to generate novel, often unexpected, hypotheses. As an example, I will focus on how the rhythmic pattern is both generated and propagated along the body during locomotion.

Eduardo Izquierdo
Fri 4:00 p.m. - 5:00 p.m. [iCal]
Panel Discussion

Author Information

Ramin Hasani (TU Wien)

Ramin is a PhD research assistant in Computer Science at the Institute of Computer Engineering, of TU Wien, Austria. His main research focus is on the development of interpretable machine learning algorithms for dynamical systems. Ramin has had his brain-inspired machine learning research featured at the TEDxVienna 2018 forum. He was a visiting research scholar at Massachusetts Institute of Technology (MIT), CSAIL, investigating interpretable neural network topologies for autonomous robotic systems. He co-chaired the 1st NIPS workshop on Worm’s Neural Information Processing (WNIP) at the 31st NIPS Conference, Long Beach, California, 2017. Ramin actively collaborates with the Distributed robotics lab at CSAIL of MIT, Institute of Science and technology (IST) Austria, Infineon Technologies, and Zimmer Group at IMP, Austria. Ramin is also a senior contributor at the OpenWorm Foundation, USA. He was a visiting scholar the Department of Computing at Imperial College London, working on developing deep learning solutions for behavioral modeling dynamic systems. Ramin has completed an M.Sc. in Electronic Engineering at Politecnico di Milano (2015) and has got his B.Sc. in Electrical Engineering – Electronics at Ferdowsi University of Mashhad (2012).

Manuel Zimmer (Research Institute of Molecular Pathology)

Since 2010 Independent group leader at the IMP, Vienna, Austria 2004 – 2010 Postdoc with Dr. Cori Bargmann, The University of California San Francisco & The Rockefeller University, New York, USA. 1998 – 2003 Ph.D. with Dr. Rüdiger Klein at EMBL-Heidelberg & Max-Planck-Institute of Neurobiology, Munich, Germany. 1998 Diploma thesis with Dr. Steven J. Burden at the Skirball Institute of Biomolecular Medicine, New York, USA. 1993 – 1998 Studies of Biochemistry at the Freie-Universität-Berlin, Germany. Selected Publications: Nichols ALA, Eichler T, Latham R, Zimmer M (2017). A global brain state underlies C. elegans sleep behavior. Science 356, eaam6851. DOI: 10.1126/science.aam6851. Hums I, Riedl J, Mende F, Kato S, Kaplan HS, Latham R, Sonntag M, Traunmüller T, and Zimmer M (2016) Regulation of two motor patterns enables the gradual adjustment of locomotion strategy in Caenorhabditis elegans. eLife 2016;5:e14116. DOI: 10.7554/eLife.14116 Kato S, Kaplan HS, Schrödel T, Skora S, Lindsay TH, Yemini E, Lockery S, Zimmer M (2015). Global Brain Dynamics Embed the Motor Command Sequence of Caenorhabditis elegans. Cell, 163(3), 656–669. DOI: 10.1016/j.cell.2015.09.034 Schrödel T, Prevedel R, Aumayr K, Zimmer M# & Vaziri A# (2013) Brain-wide 3D imaging of neuronal activity in Caenorhabditis elegans with sculpted light. Nature Methods Oct;10(10):1013-20. DOI: 10.1038/nmeth.2637. # Co-corresponding authors. Zimmer M, Gray JM, Pokala N, Chang AJ, Karow DS, Marletta MA, Hudson ML, Morton DB, Chronis N & Bargmann CI (2009). Neurons Detect Increases and Decreases in Oxygen Levels Using Distinct Guanylate Cyclases. Neuron Mar 26; 61(6): 865-879. DOI: 10.1016/j.neuron.2009.02.013 Zimmer M, Palmer A, Köhler J, Klein R (2003). EphB-ephrinB bi-directional endocytosis terminates adhesion allowing contact mediated repulsion. Nature Cell Biology Oct; 5(10): 869-878. DOI: 10.1038/ncb1045 Palmer A*, Zimmer M*, Erdmann KS, Eulenburg V, Porthin A, Heumann R, Deutsch U, Klein R. (2002) EphrinB phosphorylation and reverse signaling: Regulation by Src kinases and PTP-BL phosphatase. Molecular Cell Apr; 9(4): 725-37. DOI: 10.1016/S1097-2765(02)00488-4. * Authors with equal contribution

Stephen Larson (OpenWorm Foundation)
Tomas Kazmar (Research Institute of Molecular Pathology (IMP))
Radu Grosu (TU Wien)

Radu Grosu is a full Professor, and the Head of the Institute of Computer Engineering, at the Faculty of Informatics, of the Vienna University of Technology. Grosu is also the Head of the Cyber-Physical-Systems Group within the Institute of Computer-Engineering, and a Research Professor at the Department of Computer Science, of the State University of New York at Stony Brook, USA. The research interests of Radu Grosu include the modeling, the analysis and the control of cyber-physical systems and of biological systems. The applications focus of Radu Grosu includes distributed automotive and avionic systems, autonomous mobility, green operating systems, mobile ad-hoc networks, cardiac-cell networks, and genetic regulatory networks. Radu Grosu is the recipient of the National Science Foundation Career Award, the State University of New York Research Foundation Promising Inventor Award, the Association for Computing Machinery Service Award, and is an elected member of the International Federation for Information Processing, Working Group 2.2. Before receiving his appointment at the Vienna University of Technology, Radu Grosu was an Associate Professor in the Department of Computer Science, of the State University of New York at Stony Brook, where he co- directed the Concurrent-Systems Laboratory and co-founded the Systems-Biology Laboratory. Radu Grosu earned his doctorate (Dr.rer.nat.) in Computer Science from the Faculty of Informatics of the Technical University München, Germany. He was subsequently a Research Associate in the Department of Computer and Information Science, of the University of Pennsylvania, USA, and an Assistant Professor in the Department of Computer Science, of the State University of New York at Stony Brook, USA.

More from the Same Authors