PDP2019-01

Attractor neural networks for mapping heterogeneous spatial scales in the hippocampus

Host laboratory and collaborators

Abstract

Self-localization based on internal and/or external sensory information is essential to animal survival. The hippocampus plays a crucial role in this process by forming a mental map of the environment. The factors influencing internal spatial coding resolution are poorly understood. Recent large-scale in vivo recordings of hippocampal neuronal activity in mice navigating virtual reality environments show local variations in spatial coding resolution depending on the presence of external visual cues. They could result from genuine differences within a single internal map or possible switches between different maps associated with internal or external reference frames. To solve this issue, we propose here to use inference models based on the functional connectivity (Ising models) as a neural activity decoder to decipher at high speed the neuronal assemblies or maps in use. Map switching will be revealed as instabilities (or flickers). In a second step, we will build an attractor neuronal network model fed by both external and internal information to investigate the mechanisms of spatial resolution modulation.

Keywords

spatial representation, hippocampus, decoding, resolution, attractor networks, Ising models

Objectives

Aim1: Develop a dual decoder allowing fast decoding of cell assemblies or maps associated with different reference frames (internal vs external) and encoded positions from large-scale multi-electrode recordings of hippocampal neuron activity during spatial navigation in virtual reality environments.
Aim2: Build a Continuous Attractor Neuronal Network (CANN) model to investigate possible mechanisms involved in adaptive modulation of spatial coding resolution.

Proposed approach (experimental / theoretical / computational)

Experimental: Large-scale recordings of hippocampal cells activity in mice navigating virtual reality environments have already been recorded in the Epsztein lab and will be complemented if needed. Places within the environment differ by the availability of external visual cues (virtual objects) thus allowing navigation based on external (visual) and internal (optic flow, movements) sensory information in different parts of the same environment. Notably places near objects are coded at higher spatial resolution.

Computational/theoretical: The post-doctoral researcher will implement a statistical inference-based data analysis method. This method uses joint pairwise spiking activity as a proxy of the map used for self-location and can dissociate the map from the position coding within that map. In a second step the post-doc will develop a Continuous Attractor Neuronal Network (CANN) to explore possible underlying mechanisms for different spatial resolutions such as network instability (flickering).

Interdisciplinarity 

This project lies at the interface between neuroscience, computer science, and statistical physics. Large-scale neuronal recordings as performed in the Epsztein lab using multi-shank electrodes (up to 128 recording sites) give access to the simultaneous activity of tens to hundreds of neurons and prompt for their computational modeling. A new and promising approach relies on the inference of functional connectivity (co-activity patterns in the network) from short snapshots of recorded activity (statistical inference models based on Ising models). The recruited post-doc while working at INMED, will hold a PhD in statistical physics, mathematics or computer science. He will develop and apply an Ising model approach as developed in the Monasson/Cocco lab and close to models developed in the Rouault lab to biological questions investigated in the Epsztein lab.