Exploring Neural Principles with Si elegans, a Neuromimetic

Representation of the Nematode Caenorhabditis elegans

Axel Blau

, Frank Callaly

, Seamus Cawley

, Aedan Coffey

, Alessandro De Mauro

, Gorka Epelde

Lorenzo Ferrara

, Finn Krewer

, Carlo Liberale

, Pedro Machado

2,3

, Gregory Maclair

Thomas Martin McGinnity

2,3

, Fearghal Morgan

, Andoni Mujika

, Alexey Petrushin

Gautier Robin

and John Wade

2,3

Dept. of Neuroscience and Brain Technologies (NBT) and Nanostructures Unit (NAST),

Fondazione Istituto Italiano di Tecnologia (IIT), 16163 Genoa, Italy

Intelligent Systems Research Centre (ISRC), University of Ulster, Derry BT487JL, Londonderry, Northern Ireland

School of Science and Technology, Nottingham Trent University, Nottingham, U.K.

eHealth and Biomedical Applications (eHBA), Vicomtech-IK4, San Sebastián, Spain

Bio-Inspired Electronics and Reconfigurable Computing (BIRC), National University of Ireland, Galway, Ireland

Keywords: Brain-Inspired Computation, Nervous System Emulation, Soft Body Simulation, Virtual Embodiment,

Neurocomputational Response Models on Field-Programmable Gate Arrays (FPGAs).

Abstract: Biological neural systems are powerful, robust and highly adaptive computational entities that outperform

conventional computers in almost all aspects of sensory-motor integration. Despite dramatic progress in

information technology, there is a big performance discrepancy between artificial computational systems

and brains in seemingly simple orientation and navigation tasks. In fact, no system exists that can faithfully

reproduce the rich behavioural repertoire of the tiny worm Caenorhabditis elegans which features one of the

simplest nervous systems in nature made of 302 neurons and about 8000 connections. The Si elegans project

aims at providing this missing link. This article is sketching out the main platform components.

1 INTRODUCTION

Caenorhabditis elegans, a soil-dwelling worm with

a life span of a few days, 1 mm long and 80 µm in

diameter, is one of the five best characterized

organisms. It is multicellular and develops from a

fertilized egg to an adult worm like any other

animal. Although its genome is small (~ 10 M base

pairs), there is about 40% homology to the human

genome (3.2 G base pairs). The adult hermaphrodite

is comprised of exactly 959 cells, including 95 body

wall muscle cells and 302 neurons. Despite its

simplicity, the nervous system of C. elegans does

not only sustain vital body function, but generates a

rich variety of behavioural patterns in response to

internal and external stimuli. These include

associative and several forms of nonassociative

learning that persist over several hours (Hobert,

2003). Interestingly, many processes of learning and

memory in C. elegans are highly conserved across

evolution, which suggests that there are universal

information processing mechanisms throughout the

animal kingdom (Lin and Rankin, 2010). With all of

this data, information and modern computer

technology at hand, it is surprising that there is yet

no comprehensive artificial C. elegans emulation

system from which the principles of neural

information processing underlying behaviour can be

derived. The Si elegans project aims to fill this gap.

It will do so by implementing completely

reconfigurable neuronal models on FPGA modules

representing individual neurons. Signals between

neurons within the neural circuitry will be

exchanged by light. Such optical free-space

interconnection concept promises to be one of the

most attractive solutions for overcoming the

‘interconnectivity crisis’, currently one of the most

serious bottlenecks in upscaling neuromorphic

network architectures. In a 3D configuration, a

quasi-limitless number of connections between

modules can be established due to the fact that light

beams do not interfere with each other. It

furthermore allows the easy reconfiguration of

189

Blau A., Callaly F., Cawley S., Coffey A., de Mauro A., Epelde G., Ferrara L., Krewer F., Liberale C., Machado P., Maclair G., Martin McGinnity T., Morgan

F., Mujika A., Petrushin A., Robin G. and Wade J..

Exploring Neural Principles with Si elegans, a Neuromimetic Representation of the Nematode Caenorhabditis elegans.

DOI: 10.5220/0005190701890194

In Proceedings of the 2nd International Congress on Neurotechnology, Electronics and Informatics (-2014), pages 189-194

ISBN:

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

network connectivity by simply exchanging (or in

case of active optical elements by reprogramming)

the light distribution elements. Finally, due to the

inherent scalability of the approach, more complex

networks can be emulated.

2 THE Si elegans APPROACH

We currently develop a hardware-based computing

framework that will accurately mimic C. elegans in

real time and enable complex and realistic behaviour

to emerge through interaction with a rich, dynamic

simulation of a natural or laboratory environment.

We initiated to replicate the nervous system of C.

elegans on a highly parallel, modular, user-

programmable, reconfigurable and scalable FPGA

hardware architecture. It will be embodied in a

virtual environment for behavioural studies. The

virtualization will take the sensory-motor loop and

realistic body physics into account. The resulting

computational platform will be provided through an

open-access web portal to the scientific community

for its peer-validation and use (Figure 1).

Figure 1: The three main elements of the Si elegans

platform: 1. the hardware emulation of the C. elegans

nervous system, which is 2. virtually embodied and

interacting with an artificial environment that can 3. be

defined by users.

Several innovative key concepts will ensure the

accurate mimicry of the C. elegans nervous system

architecture and function. Each of the 302 neurons

will be represented by individual field-

programmable gate array (FPGA) modules, each of

them being independently and dynamically

programmable with a user-specific and parame-

terised neuronal response model through a user-

friendly neuron model submission and configuration

facility or through selection from a library of pre-

defined and tested neuron models. Pioneering

electro-optical interconnection schemes will allow

dense module distribution and parallel, interference-

free inter-neuron communication in a 3D space. In a

closed-loop feedback design, this hardware blueprint

of the C. elegans nervous system will control a

biophysically correct virtual representation of the

nematode body in a virtual behavioural setting.

Instead of limiting its function and impact on

science and technology by imposing pre-made

models only, the Si elegans framework will be made

available to the worldwide scientific community

through an open-access web-portal. It will feature an

intuitive and user-friendly remote configuration

interface to define an unlimited number of neuron

models and information processing hypotheses for

automatic FPGA hardware configuration. This peer-

participation concept will not only warrant the

independent and unbiased functional validation of Si

elegans, but permit the iterative optimization of

neuron models and the asymptotical approach

towards a holistic reproduction and understanding of

the complete set of C. elegans behaviours and their

underlying nervous system mechanisms through a

set of reverse-engineering tools.

Two core aspects govern the project. The first

addresses the technological design and assembly of

the Si elegans hardware architecture accompanied

by the development of the virtual arena and of the

neural response model design and FPGA

configuration interface followed by their integration

into a user-friendly web-accessible platform. The

second addresses its deployment to the scientific

community for its independent peer-validation as a

free-access tool and testbed for neurocomputational

studies.

The hardware design includes individual neuron

modules, each consisting of three elements in their

simplest embodiment as sketched out in Figure 2: i)

a synaptic/gap junction input array, ii) postsynaptic

processing of synaptic/gap junction input based on

the selected neuron-specific model that has been

implemented on dynamically (re)programmable

FPGA circuitry for the arbitrary definition of the

type of neuron and its response behaviour, and iii)

an axonal output line that distributes the neural

response (e.g., spikes, membrane fluctuations)

simultaneously to individual synapses of one or

many target neurons.

In a coarse comparison, i) represents individual

synapses/gap junctions, ii) the soma and dendritic

tree of a neuron, and iii) the axon with its axonal

arbour. The spatial organization and assembly of the

Si elegans nervous system will be based on the

NEUROTECHNIX2014-InternationalCongressonNeurotechnology,ElectronicsandInformatics

190

published connectome (taking most recent and new

findings or hypotheses published during project time

into account (e.g., (G. Haspel and O'Donovan,

2011)).

Figure 2: Concept and elements of an individual Si

elegans FPGA neuron module and its comparison with a

real neuron. Neural activity will arrive at individual input

lines of an FPGA (i) and will be processed by the neuron-

specific stimulus-response algorithm that the FPGA was

programmed with (ii). Its output activity will be

distributed in parallel through signal distribution elements

(iii) to individual input lines (i) of the target FPGA

neurons to which the neuron connects to. In case of signal

propagation by light, incoming activity will arrive as

spatially confined light pulses at individual pixels of

photoelectric converters (synapses/gap junctions) being

individually connected to the individual FPGA input lines.

Neural response activity generated by the neural model

residing on the FPGA will trigger a coherent light source

at one of its output lines (axon). This light will pass

through microoptical light-structuring elements to

distribute activity onto selected pixels (synapses/gap

junctions) of interconnected target neurons. In case of

electrical signal transmission through wires, a split-wire

bundle will transmit a digital signal pulse from the axonal

output line of the FPGA to individual synapse/gap

junction sockets of the target neurons.

2.1 FPGA Representations of

C. elegans Neurons and Muscles

The Si elegans emulation framework is composed of

distinct and independent modular components. To

date, the majority of hardware-assisted simulation

efforts have focused on mirroring a complete

network of several neurons on a single FPGA chip

or in ASIC/VLSI technology. Si elegans, in contrast,

aims at dedicating a single FPGA module to a single

neuron to emulate neuron-specific stimulus-response

models at quasi arbitrary resolution. While from an

engineering point of view this seems to be a waste of

hardware resources at first glance, this approach will

give room for implementing neural processing

schemes of arbitrary complexity allowing for the

consideration of all types of intracellular events

(e.g., signalling cascades etc.). This will allow

biologists and computational neuroscientists to

extend model complexity and fidelity to unimagined

high degrees.

By means of dynamically reconfigurable field

programmable gate arrays (drFPGAs) the inherent

signal processing and response logic of each neuron

will be reprogrammable. The need for dynamically

reprogrammable somato-dendritic circuitry has

several advantages: to stay flexible in defining the

type and thus the response behaviour of a neuron, to

implement any kind of synaptic or dendritic pre-

processing algorithms, to freely adjust or upgrade

the algorithms for emulating neural development

(changes in neural response) or implementing

upcoming neuroscience knowledge, and to emulate

disease states (e.g., Parkinson's, epileptic seizures)

by temporarily modulating the response behaviour at

run-time. drFPGAs thus offer the best ratio between

hardware costs and performance, accuracy, and

parameterization space. Since FPGA technology

currently experiences fast technological advances, it

will also be easy to exchange individual modules for

more powerful or smaller ones at any time.

2.2 3D Interconnection of FPGA

Modules to Replicate the

C. elegans Connectome

In almost all hardware implementations of neural

networks, the issue of inter-neuron connectivity is a

major problem. Serial-type simulations introduce

stochastic jitter in the timing of events and thus fail

to accurately and reproducibly mimic parallel

information flow between neurons. If a parallel

inter-neuron connectivity is implemented on-chip

instead (e.g., by using ASIC technology), typically

90% of the chip is composed of interconnect and

scaling networks becomes a major problem. In this

project we are proposing to solve this problem by

using optical or wire-based off-chip interconnects.

Two complementary interconnection strategies will

be pursued and compared for their ease of

ExploringNeuralPrincipleswithSielegans,aNeuromimeticRepresentationoftheNematodeCaenorhabditiselegans

191

implementation, reliability, functionality and

scalability. They will be implemented by adding two

elements to each FPGA neuron module: i) a

synaptic/gap junction input field (pixel matrix for

wire connection, optical fibre plugs or light-

receptive pixels with light-to-charge conversion) and

an axonal output line with distribution elements that

communicate the neural response simultaneously to

individual synapses of one or many target neurons.

For the optical information transmission among

neurons, fast-switchable and intensity tuneable laser-

diodes are the emission light source(s) of choice.

They are triggered by the axonal output pin of an

FPGA neuron module. Their light is structured and

thus projected onto individual synaptic or gap

junction inputs of those target neurons that the active

neuron connects to. Any combination of reflective,

refractive, diffractive and masking elements such as

mirrors, micromirror arrays, microprism or

microlens arrays, gratings, colour filters and etched

shadow masks will be explored to ensure the correct

addressing and optical information transfer between

sending and receiving FPGA neurons. Synapses and

gap junctions of the receiving neurons will be

represented by arrays of photoelectric elements.

2.3 A Virtual Arena for Behavioural

Studies

Several C. elegans-specific descriptors of its

physiology, morphology and body mechanics exist,

including a realistic representation of the body (e.g.,

Virtual Worm Project, neuroConstruct) and aspects

of locomotion (Boyle, 2009; Bryden and Cohen,

2008; Gal Haspel, O'Donovan, and Hart, 2010;

Mailler, Avery, Graves, and Willy, 2010; Niebur and

Erdos, 1993; Stephens, Johnson-Kerner, Bialek, and

Ryu, 2010; Wakabayashi, 2006). Based on these

data, the Si elegans hardware nervous system of C.

elegans will be embodied through a biophysically

realistic virtual representation of the nematode in a

virtual environment. The virtual body will share the

shape, body-physics (e.g., elasticity, friction) and

cellular organization of C. elegans (including

realistic spatio-functional representations of sensory

cells). The interplay between active actuation

through sensory-driven control circuits of its

nervous system and passive actuation by

environmental factors (material-properties, arena

topology, gravity, air- or fluid-flow etc.) will be

considered. The simulation of this virtual body being

situated in a virtual arena will be running on

standard PC hardware. The virtual arena can be

freely configured to copy the 3D geometries and

biophysical features of an experimental environment

used in in vivo studies. It displays the native

behaviour of the C. elegans representation, provides

simulated environmental stimuli to its sensory

neurons and shows stimuli-induced responses (e.g.,

muscle actuation, secretory events). Information

flow is channelled in real-time through a bi-

directional interface between the computer and the

Si elegans nervous system. Sensory neurons of the Si

elegans nervous system receive their input from

programmable light sources. Si elegans’ motor

neuron activity actuates associated muscles of the

virtual body. Through a real-time closed-loop

feedback, any resulting sensory experience (e.g.,

change in posture, touch, change of chemical

concentration gradients) is coded and transmitted to

the Si elegans nervous system emulation as new

sensory input.

2.4 Neural Model Definition User

Interface and Network State

Analysis

The modelling space shall allow for the definition of

relevant neural processing parameters in a pictorial,

object-oriented flow diagram (graphical drag-and-

drop manner) or script. The desired network

structure can be created by simply selecting various

neuron, synapse or gap junction models from a

library of available components and connecting

them together. Alternatively, existing neural

response models can be imported from other

simulation engines (e.g., NEURON, BRIAN,

NeuroML, ...) (Figure 3).

An assembly can be stored as a neuron-specific

model to become a high-level, properly documented

building block for other researchers. The modelling

toolset provides all required design elements to

address and freely combine all known features and

events of neural signal transmission down to the

synaptic level, possibly including even abstracted

models of signalling cascades. These design

elements can be altered, thus personalized and

versioned by community members for experimental

purposes. E.g., the function of an individual synapse

may include cable properties of the pre-synaptic

axon and synapse to account for physiological and

morphological boundary conditions that shape/affect

signal properties such as signal transmission delays

and attenuation. We furthermore implement a

standard set of amplitude-invariant, self-terminating

action potentials with stereotyped waveforms as well

as graded regenerative potentials as the predominant

signal type in C. elegans with amplitudes and

NEUROTECHNIX2014-InternationalCongressonNeurotechnology,ElectronicsandInformatics

192

waveforms that are highly sensitive to the size,

duration and waveform of the stimulus. Finally, the

Si elegans platform will provide readout and storage

tools for reverse-engineering nervous-system

function.

Figure 3: Si elegans GUI mock-up for model definition.

We will provide users with a flexible GUI for the user-

friendly and intuitive definition of neuron-specific

response models for the efficient and fast design-to-FPGA

prototyping of neural models. It will provide ready-made

neuron models and sets of parameterisable components or

functional primitives from which scientists can construct

their own neural response models. The definition

environment will contain a virtual representation of C.

elegans including the locations and connections of

individual neurons and muscle cells (and everything else).

The user may zoom into the nervous system and click on a

neuron, which will then show its incoming and outgoing

connections within the nervous system along with their

morphologically correct pathways from the source neurons

(incoming) towards the target neurons/cells (outgoing).

Upon double-click of elements in the GUI a dialog will

open for defining the properties of individual input-output

algorithms (neural response model - e.g., ion-channel

response, synaptic integration etc.). A CAD-program-like

tree lists all features/synapses/primitives of the particular

neuron. A separate tab allows the definition of information

flow among individual elements (blocks) within a

particular neuron (e.g., dendro-somatic kinetics, leakage

…). Alternatively, these parameters may also be imported

from other simulation engines or be defined by a

Python/PyNN definition script.

2.5 Deployment of the Si elegans

Platform for Public Access and

Peer-Contribution

Given that there are over 300 laboratories worldwide

working on C. elegans, a majority of them studying

some aspect of nervous system function, there is

considerable interest in modelling it. The seemingly

trivial complexity of the nematode C. elegans has

kept researches busy over the past 50 years without

revealing a complete understanding of the

functioning of its nervous system. The spirit and

central mission of Si elegans as an ‘open-science,

peer-contribution’ project is therefore the early

involvement of the scientific community,

particularly groups interested in behavioural and

modelling aspects of C. elegans, but also the neuro-

computation community at large. The Si elegans

platform will therefore be made accessible to the

scientific community (e.g., in a shared-time, fair-use

model) through a client/server-based remote access

environment not only for the pursuit of scientific

studies, but also for its independent validation and

enhancement through extensions (e.g., through plug-

ins). Feedback from researchers will be shared

through community outreach tools and be

implemented into the public platform (Figure 4).

Figure 4: Schematic overview on the Si elegans platform

elements and their accessibility by the scientific

community.

3 CONCLUSIONS

While Si elegans restricts itself to the emulation of

the C. elegans nervous system, the underlying

design concepts have universal application. Si

elegans will constitute a generalizable framework

from which the universal working principles of

nervous system function can be induced, and new

scientific knowledge on higher brain function and

behaviour can be generated. By being designed for

open-source and peer-contribution, Si elegans

promises to become a generalizable testbed for

neurobiological signal processing hypotheses in

healthy and disease-affected nervous systems in

neuroscience, neurocomputation and neurology. It

may also lay the foundation for exploring and

refining new neuromimetic computational concepts

ExploringNeuralPrincipleswithSielegans,aNeuromimeticRepresentationoftheNematodeCaenorhabditiselegans

193

to provide a blueprint for the design of biologically

inspired, brain-like parallel processing hardware

architectures that are orthogonal to current von

Neumann-type machines. If a holistic and verifiable

understanding of C. elegans nervous system function

could be achieved, a long-locked door would open to

implement such knowledge in neuromimetic

processing and control mechanisms in any

technological field, be it robotics, medical

assistance, decision-making devices, fraud-detection

or surveillance.

ACKNOWLEDGEMENTS

The Si elegans project 601215 is funded by the 7

Framework Programme (FP7) of the European

Union under FET Proactive, call ICT-2011.9.11:

Neuro-Bio-Inspired Systems (NBIS).

REFERENCES

Boyle, J. H. (2009). C. elegans locomotion: an integrated

approach. (PhD Thesis), University of Leeds, Unitied

Kingdom. Retrieved from http://etheses.whiterose.

ac.uk/1377/

Bryden, J., and Cohen, N. (2008). Neural control of

Caenorhabditis elegans forward locomotion: the role

of sensory feedback. Biological Cybernetics, 98(4),

339-351. doi: 10.1007/s00422-008-0212-6

Haspel, G., and O'Donovan, M. J. (2011). A perimotor

framework reveals functional segmentation in the

motoneuronal network controlling locomotion in

Caenorhabditis elegans. Journal of Neuroscience,

31(41), 14611-14623. doi: 10.1523/JNEUROSCI.

2186-11.2011

Haspel, G., O'Donovan, M. J., and Hart, A. C. (2010).

Motoneurons Dedicated to Either Forward or

Backward Locomotion in the Nematode Caenor-

habditis elegans. The Journal of Neuroscience, 30(33),

11151-11156. doi: 10.1523/jneurosci.2244-10.2010

Hobert, O. (2003). Behavioral plasticity in C. elegans:

Paradigms, circuits, genes. Journal of Neurobiology,

54(1), 203-223. doi: 10.1002/neu.10168

Lin, C. H., and Rankin, C. H. (2010). Nematode Learning

and Memory: Neuroethology. In M. D. Breed and J.

Moore (Eds.), Encyclopedia of Animal Behavior (pp.

520-526). Oxford: Academic Press.

Mailler, R., Avery, J., Graves, J., and Willy, N. (2010). A

Biologically Accurate 3D Model of the Locomotion of

Caenorhabditis Elegans. Paper presented at the 2010

International Conference on Biosciences

(BIOSCIENCESWORLD).

Niebur, E., and Erdos, P. (1993). Theory of the

locomotion of nematodes: control of the somatic

motor neurons by interneurons. Mathematical

Biosciences, 118(1), 51-82.

Stephens, G. J., Johnson-Kerner, B., Bialek, W., and Ryu,

W. S. (2010). From Modes to Movement in the

Behavior of Caenorhabditis elegans. PloS One, 5(11),

e13914. doi: 10.1371/journal.pone.0013914

Wakabayashi, M. (2006). Computational Plausibility of

Stretch Receptors as the Basis for Motor Control in C.

elegans. (BA Thesis), Univ. of Queensland, Australia.

Retrieved from http://archive.itee.uq.edu.au/~markw/

celegans/thesis.pdf?origin=publication_detail

NEUROTECHNIX2014-InternationalCongressonNeurotechnology,ElectronicsandInformatics

194