Models for translation and cross-species bridging

Quantify behaviour and predict neuronal firing modes

Resource Mobilisation

In this workspace-overarching project, we use computational modelling to build bridging principles for transferring our mechanistic understanding in rodents to non-human primates and, further, to human applications (in collaboration with the DECODE platform).
First, we will establish common approaches to be used in all projects to quantify cross-species behaviour. Recently developed methods for video analysis, especially models for pose estimation and behaviour classification, are transforming behavioural quantification to be more precise, scalable, and reproducible. Through our previous work, we have overcome long-standing limitations of manual scoring of video frames. We have already applied several approaches to video-tracking using optical flow. Our developed algorithm VAME uses self-supervised deep learning models to infer the full range of behavioural dynamics based on the animal movements from pose data (using DLC or comparable approaches). The variational autoencoder framework is used to learn a generative model. An encoder network learns a representation from the original data space into a latent space. A decoder network learns to decode samples from this space back into the original data space. The encoder and decoder are parameterized with recurrent neural networks. Once trained, the learned latent space is parameterized by a Hidden Markov Model to obtain behavioural motifs.
Second, we will build full-morphology compartmental models of individual neurons and simple circuit-level models to predict thermal and mechanical effects on neuronal input-output transformations. These models integrate ion channel properties, dendritic fiber properties and synaptic conductances and membrane electrical properties (e.g. Piezo-channels). They will inform us about the expected temperature dependence of neuronal firing mode transitions and synaptic plasticity (engram) formation. In addition, we will build models that allow for a tuning of stimulus strength based on tissue heating and propagation, that will be confirmed by MR-based thermometry approaches in rodents (9.4 T small animal scanner) and humans. Our aim is to extrapolate rodent data to increasing scales to predict effects and to improve invasive and non-invasive neurostimulation applications in non-human primates and humans.