Project background and motivation:

A key aspiration of brain research is to understand the mechanisms allowing animals to translate environmental cues (inputs) into distinct behavioural decisions (outputs). How does the underlying neuronal network sense environmental signals? How is information processed and integrated in neuronal networks to drive specific animal decisions? A powerful approach for characterizing neuronal functions is perturbation experiments: quantifying how inhibition of neurons affects input processing and behavioural outputs execution. Yet, this approach is challenging due to the inherent complexity and scale of the brain (~10¹¹ neurons).

In our research, we study sensation and decision making using the compact nervous system of C. elegans worms that have only 302 (!) neurons. See Experimental Set-up Demo For that, we developed microfluidic devices, imaging tools, and genetic engineering techniques that allow us to: (1) monitor animal behavioural responses to various controlled environmental cues for hundreds of animals per day. (2) acutely inhibit single neurons of our choice.

Project description

This project uses data from high-throughput behaviour experiments to decipher the function of neurons in sensation and decision making. Students will:

1. Develop an automated analysis for quantifying behaviour in response to the controlled cues delivered in microfluidic devices. Typical cues (inputs) are the concentration of odorants (‘food cues’) or glycerol (osmotic stress). Typical behaviour outputs are speed and angular velocity profiles, and probability distributions of worm navigation behaviours (e.g., forward, reverse, pause, stereotypical ‘omega’, sharp turns, and curving). Large datasets are involved - thousands of animals, variety of behaviour features, various cues, and different environmental contexts.
2. Identify significant input-output correlations and context dependencies.
3. Analyse behaviour of genetically engineered animals in which specific neurons are inhibited.
4. Extract the function of neurons in sensation and decision making by comparing the behaviour of animals with perturbed neurons (section 3) with the behaviour of normal animals (wild type).

The project is offered for one semester and can be extended to two semesters.

Prerequisites:    - Highly motivated students with passion for Neuroscience and Biology

- Experience with programming in Matlab/Python (Scripts will be written in Matlab).

- Knowledge in image processing and/or signal processing.

Contact Details:    Sagi Levy  levysagi@technion.ac.il  , Biology Faculty

https://sagilevy.net.technion.ac.il/research/

Project background and motivation:

How do neuronal networks encode a perceived signal or an internal state? How does information flow within neuronal networks, and which information is encoded in their dynamics? How do the neurons’ noise, delay and non-linearity impact information processing?

A valuable approach for studying information processing in neuronal networks is by simultaneously monitoring neuronal activity of large neuronal populations, and analyzing their activity features. In two model organisms, larval zebrafish and C. elegans worms, activity can be simultaneously recorded in all neurons in the brain at a single-cell resolution, allowing unbiased analysis of comprehensive neuronal networks. Whole brain imaging data is typically analyzed at the neuron population level, revealing invaluable information about neuronal dynamics, neuronal network organization and function (e.g. ¹). Yet, a key challenge in whole-brain imaging is to uniquely identify each neuron. In the absence of precise and systematic identification tools, it is difficult to monitor the same cell across animals, limiting the capacity to discover single-neuron functions. The challenge of unique neuronal identification was very recently addressed in C. elegans worms by designing a multicolor strain in which proximal neurons are differentially labeled (NeuroPAL²). In contrast to the previous Brainbow cell labeling technique³, NeuroPAL colors are not stochastic, allowing unique neuronal identification over time and across animals.

In our research, we study how information is processed in neuronal networks by imaging the activity of all neurons in C. elegans worms head, while monitoring their cellular identity across animals (NeuroPAL²). We record neuronal activity (outputs) in response to controlled stimuli (inputs) precisely delivered using specialized microfluidic devices.

Project description

This project involves image analysis of whole-brain imaging videos, extraction of neuronal activity and identity, and meta-analysis for identification of neuronal network features. Scripts will be developed in Matlab. Students will:

1. Develop image analysis tools for whole brain-imaging videos (timelapse confocal microscopy with multi-channel imaging at 3D), including neuron segmentation, neuron identification (NeuroPAL) and extraction of neuronal activity dynamics.
2. Analyze mutual information between different neurons, and between neurons and input stimuli (concentrations of an odorant “food cue” or glycerol osmotic stress).
3. Identify features underlying information processing from meta-analysis of imaging experiments

The project is offered for one semester and can be extended to two semesters.

Prerequisites:    - Highly motivated students with passion for Neuroscience and Biology

- Experience with programming in Matlab/Python (scripts will be written in Matlab).

- Knowledge in image processing.

Contact Details:    Sagi Levy  levysagi@technion.ac.il  , Biology Faculty

https://sagilevy.net.technion.ac.il/research/

1      Kato, S., Kaplan, H. S., Schrodel, T., Skora, S., Lindsay, T. H., Yemini, E., Lockery, S. & Zimmer, M. Global brain dynamics embed the motor command sequence of Caenorhabditis elegans. Cell 163, 656-669 (2015).

2      Yemini, E., Lin, A., Nejatbakhsh, A., Varol, E., Sun, R., Mena, G. E., Samuel, A. D. T., Paninski, L., Venkatachalam, V. & Hobert, O. NeuroPAL: A Multicolor Atlas for Whole-Brain Neuronal Identification in C. elegans. Cell 184, 272-288 e211 (2021).

3      Weissman, T. A. & Pan, Y. A. Brainbow: new resources and emerging biological applications for multicolor genetic labeling and analysis. Genetics 199, 293-306 (2015).

Research proposal - Abstract:

Currently, Deep learning models show remarkable results on computer vision tasks by optimizing the network parameters, such as network depth, layers connections, weights initialization, and network hyperparameters.

A fascinating observation is that biological systems fail differently than man-made machines as they age. In this work, we will study how these Deep networks perform as a function of age-associated failure. We will explore the effect of aging of different state-of-the-art networks, such as Vgg16, Resnet50, Unet, Efficientnet as various datasets such as ImageNet, CIFAR10, and CIFAR100.

Project prerequisites:

• Deep Learning course - must
• Matlab/Python code base

Setup:

• Deep Computer with GPU, pycharm/Matlab IDE

Supervision:

• Nati Daniel – 0544451026, snatidaniel@gmail.com
• SavirLab, Rappaport Faculty of Medicine, Technion, https://savir.net.technion.ac.il/