AI Opportunities for Obsessive-Compulsive Disorder. Q&A with Henry Szechtman
“This vast number of records allows utilisation of Machine Learning/Artificial Intelligence algorithms.”
A study published in GigaScience provides an enormous amount of behavioural science data, presented in a detailed videographic virtual library, that was used to explore Obsessive-Compulsive Disorder (OCD) in an animal model. The virtual library comprises 11.1 TB of videography behavioural science data from a 15-year study and includes over 2 years of continuous recording (see EurekAlert Press Release).
For a GigaBlog Q&A, I discussed with principal investigator Henry Szechtman (McMaster University) his core interest in the project, and the immense reuse potential of these data.
What inspired your interest in Obsessive-Compulsive Disorder?
“Actually, I was led to the study of Obsessive-Compulsive Disorder by the results of my research. Since graduate studies at the University of Pittsburgh I was interested in the role of dopamine systems in normal and abnormal behaviour, which led me to analyse the activity of rats administered dopamine stimulant drugs. In one such study, we analysed the activity of rats injected with the dopamine receptors stimulant drug, apomorphine. I begun this study in Israel, where I was a Weizmann Fellow, and was fortunate to team up with Drs Ilan Golani (Tel-Aviv University) and Philip Teitelbaum (University of Illinois) who were collaborating and applying the Eshkol-Wachman Movement Notation to analyse behaviour of rats with lesions to the lateral hypothalamus.”
“I was interested to learn and apply the same technology to study apomorphine-induced behaviour, and fortunately for me Drs Golani and Teitelbaum found such a project of interest too. This project introduced me not only to a method of behavioural analysis but also to the importance of having a visual record of behaviour as rat activity was being analysed frame-by-frame from 16 mm films (video recording technology became available later).”
“How did I go from apomorphine-induced behaviour to an interest and research of OCD? Our analysis of apomorphine-induced behaviour was in the context of investigating the then popular dopamine hypothesis of schizophrenia. However, when we switched to quinpirole, a drug that was selective to only a subtype of dopamine receptors stimulated by apomorphine, our behavioural analysis did not show behaviour that was disorganised, as expected from an animal model of schizophrenia.”
“Obsessive-Compulsive Disorder is a disturbance of a vital Security Motivation System that manages the behavioural response to biologically relevant potential threats”
“A young psychiatrist at a departmental seminar I presented suggested that quinpirole-induced behaviour was reminiscent of OCD, a psychiatric condition I had not heard of before but that was just becoming familiar to the general public. Indeed, as David Eilam (Tel-Aviv University), Bill Sulis (McMaster University) and I examined the behaviour of quinpirole-induced behaviour in detail, we were able to propose that the transformation in behaviour induced by quinpirole had the attributes of an animal model of OCD and of compulsive checking in particular.”
“The work on OCD led Erik Woody (University of Waterloo) and me to propose in 2004 a relatively new theory that OCD is a disturbance of a vital Security Motivation System that manages the behavioural response to biologically relevant potential threats.”
In the GigaScience manuscript, you highlight the reuse potential of the videography Virtual Library data “as inputs for new experiments or for yet unknown types of analysis of behaviour”. Can you expand on how one should address this challenge?
“The Virtual Library can serve as input to new experiments because of the annotation schema, which specifies for each trial a comprehensive list of independent variables and thereby renders each open field test as an independent event. Hence, choosing from the list of independent variables a judicious set for the experimental and control groups, one constructs in essence a new experimental design, outside any published study performed by the original investigators.”
“To give a simple example: If the experimental question were whether compulsive checking (or any other dependent variable) is more intense after 8 injections of quinpirole or 8OH-DPAT or mCPP, then one would select from the pool of 19,978 trials all the tests in the open field where a rat without lesion received 8 injections of quinpirole or 8OH-DPAT or mCPP or saline (the control injection) and compared their response. However, because in the Virtual Library each of those drugs was administered at several doses, selection on an additional criterion is necessary, namely, equivalent doses across drugs. If enough trials meeting those criteria are present in the Virtual Library, then those are the inputs for a new experiment.”
“The re-use potential of the Virtual Library is even greater because of videography: the Virtual Library contains not only the digitised coordinates of locomotion in the open field (from which measures of compulsive checking and the amount of activity can be derived) but also the video record of rat’s activity in the open field. These video recordings constitute the full documentation of the animal behaviour in the predefined environment. As such, they have information beyond the paths of locomotion that were of interest in our experiments.”
“Generally, the re-use potential of videography experiments is not fully appreciated.”
“For instance, the videos contain information as to what other movements (aside from locomotion) the rat performs, what are their characteristics, and what are the salient features of ritualistic behaviour, etc. Moreover, the viewer, by watching the rat’s behaviour in the videos, may formulate questions/hypotheses regarding the rat’s activity that did not occur to the experimenter prior to doing the study. Similarly, having video recordings of the experiment may allow the investigator to perform in the future analyses of behaviour that were not available at the time of the study.”
“Generally, the re-use potential of videography experiments is not fully appreciated. Many commercial software tracking systems can track the animal directly in real time, rather than offline from video records. Although a video of the trial can be acquired concurrently, most investigators do not use this option. Regrettably, that is a missed opportunity to contribute to building a virtual library of open linked datasets of behavioural performance in standard conditions.”
Are there any specific questions that utilise Machine Learning / Artificial Intelligence that should be explored on your videography Virtual Library?
“The library contains videos and quantitative recordings of the animal locations in time. The recordings of locations can be processed by the user to allow extraction of various information based on the timestamps and the coordinates of the animal. Each 55-minute session yields a sequence of more than 95,000 records of locations of the animal, and each session can be examined either separately from the others or as a part of a sequence of sessions the animal was exposed to.”
“This vast number of records allows utilisation of Machine Learning/Artificial Intelligence algorithms. For example, these algorithms can be utilised to predict the escalation of compulsive checking in animals treated with different drugs, and the decline in exploratory behaviour of control animals, as well as to identify the range of unique patterns of escalation and decline across individual rats and types of treatments.”
“Other questions can involve examination and prediction of activity patterns; evolution of the places of focus in the environment during development of compulsive checking or as a function of different treatments; identification of outlier cases and predictor variables.”
“…by learning from an animal model of OCD, a Machine Learning/Artificial Intelligence algorithm would become a model that behaves like an organism with OCD.”
“Moreover, Machine Learning/Artificial Intelligence algorithms can be used to identify measures of behaviour with long-term dependencies across years or with seasonal periodicity. Importantly, one could analyse patterns of recurrence in behaviour and teach algorithms to capture that intrinsic variability so that they can then become virtual generative processes of behaviour. In other words, by learning from an animal model of OCD, a Machine Learning/Artificial Intelligence algorithm would become a model that behaves like an organism with OCD.”
Read the GigaScience Article:
Henry Szechtman, Anna Dvorkin-Gheva, Alex Gomez-Marin, A virtual library for behavioral performance in standard conditions—rodent spontaneous activity in an open field during repeated testing and after treatment with drugs or brain lesions, GigaScience, Volume 11, 2022, giac092, https://doi.org/10.1093/gigascience/giac092
