COMPUTATIONAL PSYCHIATRY COURSE 2021
13th September 2021 - 18th September 2021
This course is designed to provide students across fields (neuroscience, psychiatry, physics, biology, psychology....) with the necessary toolkit to master challenges in computational psychiatry research.
The CPC is meant to be practically useful for students at all levels (MDs, Master, PhD, Postdoc, PI) coming from diverse backgrounds (neuroscience, psychology, medicine, engineering, physics, etc.), who would like to apply modeling techniques to study learning, decision-making or brain physiology in patients with psychiatric disorders. The course will teach not only the theory of computational modeling, but also demonstrate software in application to example data sets.
The goal of the Computational Psychiatry Course (CPC) is to create a scientific and educational space for students, scientists, and other professionals to share and advance the state of knowledge in CP. Everyone is welcome at the CPC.
To this end, we encourage all participants to treat each other respectfully. This code of conduct defines a set of guidelines to facilitate this.
Detailed information and more material can be found on the course website.
The CPC is divided into two parts: The main course (Day 1-5) and in-depth practical tutorials (Day 6).
The first day will cover topics in Psychiatry providing a conceptual basis for the type of questions that Computational Psychiatry will need to address.
The second day will explain basic modelling principles (basic mathematical terminology, step-by-step guide on how to build a model, model fitting and model selection) and will finish with a first introduction to models of perception (Psychophysics, Bayesian Models od Perception).
The third day will continue with reinforcement learning, models of perception (Predicitve Coding), action selection (Markov Decision Processes, Active Inference, Drift Diffusion Models) and will end with an introduction to models of metacognition
The fourth day will include models of connectivity (Dynamic Causal Modeling for fMRI and EEG and biophysical network models) and Machine Learning (basics and advanced).
The fifth day will feature a series of talks on practical applications of computational models to problems from psychiatry.
The practical tutorials on the sixth day will provide 3-hour, small-group, in-depth and hands-on sessions on a specific modelling approach. The practical sessions cover only open-source software packages. The code can be found under the respective links below. If you sign up, you will receive an installation guide and further information before the course takes place.
- Practical Session A with Tore Erdmann, Alexander Hess & Lilian Weber
Bayesian Learning using the Hierarchical Gaussian Filter (HGF, TNU Tapas)
In this tutorial, we will recap the theory behind the Hierarchical Gaussian Filter (HGF) and introduce the model in an accessible way. We will then discuss practical issues when fitting computational models to behavioral data in general and specific to the HGF. We will work through exercises to learn how to analyze data with the HGF using the HGF-toolbox (in Matlab).
- Practical Session B with Conor Heins & Daphne Demekas
Active Inference using the PYMDP Toolbox
This tutorial provides a practical guide on developing computational models using `pymdp`, a Python package for solving partially-observed Markov Decision Processes (POMDPs) with Active Inference. Students will build simple simulations in interactive, cloud-hosted Python notebooks (Google Colab). We aim to help students build generative models for POMDPs and to develop a conceptual understanding of the theoretical principles behind active inference, without requiring detailed technical knowledge.
- Practical Session C with Woo-Young Ahn, Mina Kwon & Hoyoung Doh
Reinforcement Learning using the hBayesDM Package
In this tutorial, participants will learn how to use a Bayesian package called hBayesDM for modeling various reinforcement learning and decision making (RLDM) tasks. A short overview of (hierarchical) Bayesian modeling will be also provided. Participants will also learn important steps and issues to check when reporting modeling results in publications.
- Practical Session D with Mads Lund Pedersen
Drift Diffusion Models using the HDDM Toolbox
The tutorial will provide a practical introduction to analyzing decision making data with the drift diffusion model (DDM) using the open source python toolbox HDDM. We will go through practical steps of applying the models to data, from importing data to running and validating models. We will also briefly go over extensions to the DDM, including the reinforcement learning drift diffusion model (RLDDM).
- Practical Session E with Lionel Rigoux & Matthias Müller-Schrader
Model Inversion using the Variational Bayes Toolbox
This hands-on tutorial is a crash course on practical computational modelling. You will build a simple model (delay discounting) and learn how to apply it on empirical data to perform parameter estimation and model selection. We will use the VBA-toolbox which contains all the tools to simulate, estimate, and diagnose your models, as well as a collection of ready-to-use models (eg. Q-learning, DCM). No previous experience with modelling is required, but basic knowledge of Matlab is recommended.
- Practical Session F with Thomas Wolfers & Saige Rutherford
Machine Learning using the PCNtoolkit
Would you like to learn more about modeling individual differences and heterogeneity in psychiatry? In this tutorial, we will abandon the classical patient vs. healthy control framework. You will be guided through how to run an analysis using normative modeling implemented in the PCNtoolkit.
- Practical Session G with Rosalyn Moran
Dynamic Causal Modelling for EEG using the SPM-DCM Package
This tutorial will examine specific features of EEG data that can be used to optimize a cell and receptor specific model of brain connectivity. EEG data acquired from event related (ERP) and temporally extended / resting state studies (Spectral responses) will be examined. The neural mass models - their assumptions and parametrizations will be compared. Students will learn to use the SPM graphical user interface and to write batch code in MATLAB to perform Dynamic Causal Modeling of EEG.
- Practical Session H with Jakob Heinzle & Birte Toussaint
Dynamic Causal Modelling for fMRI using the SPM-DCM Package
In this tutorial you will learn how to use the SPM software to perform a dynamic causal modeling (DCM) analysis in MATLAB. We will first guide you through all steps of a basic DCM analysis of a single subject: Data extraction, Model setup, Model inversion and, finally, inspection of Results. We will then proceed to look at a group of subjects. Here, we will focus on model comparison and inspection of model parameters.
We will provide a point-by-point recipe on how to perform the analysis. However, it is of advantage if you have worked with neuroimaging (fMRI) data and MATLAB before.
- Practical Session I with Marion Rouault & Ashraya Indrakanti
Metacognition using the hMeta-d Toolbox
In this tutorial, we will recap the theory underlying the hMeta-d model for quantifying metacognitive efficiency, our ability to monitor and evaluate our own decisions. We will introduce the model in an accessible way, then discuss practical issues when fitting computational models to behavioral data and go through code examples using the hMeta-d toolbox.
- Practical Session J with Stefan Frässle
Advanced Models of Connectivity: rDCM using Tapas rDCM
In this tutorial, you will learn how to use the regression dynamic causal modeling (rDCM) toolbox to perform effective (directed) connectivity analyses in whole-brain networks from functional magnetic resonance imaging (fMRI) data. We will provide you with the necessary theoretical background of the rDCM approach and detail practical aspects that are relevant for whole-brain connectivity analyses. After having laid the foundation, a hands-on part will allow you to obtain a better feeling for the behavior of the model as well as provide you with in-depth experience on how to apply the model to empirical fMRI data. The goal of this tutorial is to familiarize you with the theoretical and practical aspects of rDCM, which will allow you to seamlessly integrate the approach into your own research. We will provide clear instructions on how to perform the analyses. However, experience with the analysis of fMRI data (already some experience with classical DCM for fMRI would be ideal) as well as experience with MATLAB are beneficial.
Woo-Young Ahn, Seoul National University, South Korea
Daphne Demekas, Imperial College London, United Kingdom
Hoyoung Doh, Seoul National university, South Korea
Tore Erdmann, Scuola Internazionale Superiore di Studi Avanzati, Italy
Michele Ferrante, National Institute of Mental Health (NIMH), USA
Stefan Frässle, University of Zurich & ETH Zurich, Switzerland
Conor Heins, Max Planck Institute of Animal Behavior and University of Konstanz, Germany
Jakob Heinzle, University of Zurich & ETH Zurich, Switzerland
Marcus Herdener, University of Zurich, Switzerland
Alex Hess, University of Zurich & ETH Zurich, Switzerland
Quentin Huys, Max Planck UCL Centre for Computational Psychiatry and Ageing Research, United Kingdom
Ashraya Indrakanti, University of Zurich & ETH Zurich, Switzerland
Mina Kwon, Seoul National University, South Korea
Roland von Känel, University Hospital Zurich, Switzerland
Andre Marquand, Donders Institute, Netherlands
Christoph Mathys, Aarhus University, Denmark
Rosalyn Moran, King's College London, United Kingdom
Graham Murray, University of Cambridge, United Kingdom
John Murray, Yale School of Medicine, USA
Matthias Müller-Schrader, University of Zurich & ETH Zurich, Switzerland
Matthew Nassar, Brown University, USA
Yael Niv, Princeton University, USA
Gina Paolini, Klinik für Psychiatrie und Psychotherapie, Clienia Schlössli AG, Switzerland
Mads Lund Pedersen, University of Oslo, Norway
Inês Pereira, University of Zurich & ETH Zurich, Switzerland
Frederike Petzschner, Brown University, USA
Lionel Rigoux, Max Planck Institute for Metabolism Research Cologne, Germany
Jonathan Roiser, University College London, United Kingdom
Marion Rouault, École Normale Supérieure, France
Saige Rutherford, Donders Institute, Netherlands
Liane Schmaal, University of Melbourne, Australia
Helen Schmidt, University of Zurich & ETH Zurich, Switzerland
Jakob Siemerkus, University of Zurich & ETH Zurich, Switzerland
Ryan Smith, Laureate Institute for Brain Research, USA
Klaas Enno Stephan, University of Zurich & ETH Zurich, Switzerland
Birte Toussaint, University of Zurich & ETH Zurich, Switzerland
Lilian Weber, University of Zurich & ETH Zurich, Switzerlan
Thomas Wolfers, Donders Institute, Netherlands
Angela Yu, University of California, USA
Translational Neuromodeling Unit
University & ETH Zurich