Job ID: 106510
PhD project – An efficient modelling approach for the detection of spiking motifs in neurobiological data
Position: Ph.D. Student
Deadline: 2 April 2023
Employment Start Date: 2 October 2023
Contract Length: 3 years
Institution: Aix Marseille Université
The NeuroSchool PhD Program of Aix-Marseille University (France) has launched its annual calls for PhD scholarships for students with a master’s degree in a non-French university.
The following project is one of the 14 proposed projects. Not all proposed projects will be funded, check our website for details.
State of the art. Neuroscience has recently undergone a scientific and technological revolution. The scales at which neuronal activity can be experimentally recorded has considerably expanded. One striking example is the use of photonic imaging approaches to simultaneously sample the activity of thousands of neurons in vivo. Such novel experimental evidence shows that information processing in the brain is not a purely feed-forward process but relies also on internally generated activity in recurrent networks forming complex dynamical systems. Interestingly, it has been recently shown that neural information can be carried by way of series of spikes distributed on neurons of large networks and forming precise spiking spatio-temporal motifs.
Objectives. The goal of this project is to bring an interdisciplinary perspective to the detection of precise spike motifs in neurobiological data. In particular, inspired by neurobiological observations, we will mathematically formalize a representation in an assembly of neurons based on a set of motifs with different relative spike times. The main innovative aspect is to consider a representation based on repetitions of these spiking motifs at precise times of occurrence, thus extending the capabilities of analog representations based on vectors of instantaneous firing rate.
Methods. In preliminary work, we showed that this dedicated artificial neural network outperforms classical covariance-based methods in recognizing spiking motif timing and identity. This machine learning algorithm is particularly powerful when there is a large number of overlapping motifs as the temporal depth of the motifs increases. We used the pyTorch deep learning library, which is well suited for high-performance computing architectures such as GPUs, making it a viable option for high-throughput analysis of neurobiological data.
Expected results. An added value of this algorithm is that it can be used to learn precise spiking motifs in an unsupervised manner. It is a powerful tool for the detection of sequential activation of motifs in neurobiological data. In particular, the motifs may take the form of elementary waves and we will challenge the hypothesis that neural activity drives computations thanks to traveling waves, an aspect in which the collaboration with the team of Frédéric Chavane, an expert on their functional role in the visual cortex, will be crucial.
Feasibility. Thanks to our preliminary work, we have validated this algorithm on synthetic data for which the ground truth is known. The method will be applied on existing data by our group, notably from cats, macaques or marmosets. The availability of this material makes the project highly feasible.
Expected candidate profile, specifying at least 4 skills. The candidate should have a strong inter-disciplinary profile in 1/ computational neuroscience, 2/ biological neuroscience, 3/ machine learning, and 4/ open science (FAIR principles).