Description
The TramWay project aims at designing an entirely automated procedure to study the random walks (RWs) of biomolecules in living cells. At the nanoscale, the dynamics of individual biomolecules is inherently random, governed by thermal noise and stochastic molecular interactions. By giving access to the full distribution of molecular properties, rather than simply their average value, the great advantage of single molecule measurements is thus their ability to identify static and dynamic heterogeneities, and rare behaviours. Local heterogeneities, numerous specific and non-specific interactions, and unknown number of molecular partners in the cellular environment lead to extremely unusual properties of biological RWs, making the inference of their nature a challenging statistical and computational problem.
More information on the application of TRamWAy to synapses is available at the following address: https://research.pasteur.fr/en/project/mapping-receptor-dynamics-in-synapses/
We are proposing a 2-year postdoctoral fellowship financed by the “TRamWAy” ANR project for a motivated applicant. She/he is expected to contribute to our efforts in designing amortized inference approaches for RWs described by predictive models and in developing an unsupervised structured inference framework to probe random walks that cannot be described by a canonical model, e.g. such as a fractional Brownian motion or a hidden Markov model. We are looking for either a physicist with a strong statistical physics background or a computer scientist with experience in complex, large-scale variational inferences.
The successful candidate will have privileged access to multiple GPU clusters in the Pasteur institute as well as a CPU cluster. Furthermore, she/he will join a full team working on the project composed of Christian L. Vestergaard, who developed a non-tracking algorithm using belief propagation and unseen graph summation to infer biomolecule dynamics at high density, Francois Laurent, who designed the TRamWAy software platform and is currently developing stochastic gradient approaches to infer time varying 2D/3D maps of biomolecule dynamics at the full cell scale, and Alexander Serov, who after designing Bayesian evidence tests for Ito-Stratonovitch dilemma is developing amortized approaches to detect out-of-equilibrium dynamics from recorded RWs.
Required Skills
For a physicist: statistical physics and random walk theory
For a computer scientist: Bayesian Inference, structured and variational inferences
Scripting language experience: Bash, Python
Nice to have Skills
Tensorflow/Keras/pytorch
Duration
2 years
Localization
Institut Pasteur (Paris)
Contact Information
Interested candidates should contact Jean-Baptiste Masson (jbmasson@pasteur.fr) with their CV, a brief motivation letter and 2 letters of recommendation.