Povilas Karvelis, PhD

Neurocomputational models for predicting suicidality in major depression

Suicide is a global phenomenon and occurs throughout the lifespan. The majority of annual suicides are estimated to occur during a depressive episode, with major depressive disorder (MDD) patients having a 20-fold risk of dying by suicide than the general population. Despite decades of research seeking to identify and treat risk factors, medicine lacks non-invasive, predictive tests for accurately predicting individual risk for suicidality and prognostic trajectories of suicidal thoughts and behaviours among individuals suffering from MDD. To overcome this impasse, we craft cognitive tasks assessing suicidal ideation in MDD, quantify the underlying aberrant inference mechanisms using mathematical modelling of behaviour, link these mechanisms to specific neurocircuitry by fMRI measurements performed during the designed task, and test the clinical utility of this framework in a cross-sectional and prospective study of MDD patients with suicidal ideation.

Perceptual inference with continuous variables, continuous time and complex priors

Bayesian accounts of schizophrenia and autism characterize these disorders as an imbalance in precisions that weigh sensory likelihoods and top-down priors. Most of the recent task designs and computational models investigating this hypothesis, however, are limited to inference on binary variables and/or unimodal priors. Furthermore, inference on each trial is often treated as a durationless instant, ignoring within-trial dynamics of evidence accumulation. We have developed a model that can address all of these limitations: Continuous Response Scale Drift-diffusion Model (CDM) supports inference in continuous time, on continuous variables and with complex (non-unimodal) priors. CDM is an extension of the classical 2-dimenional Drift-diffusion model (DDM) to 3 dimensions. The model is validated via simulations and by fitting it to two datasets (autistic traits and clinical schizophrenia) that were previously analyzed using a simple Bayesian model which ignored within-trial dynamics.

Valton, V. *, Karvelis, P. *, Richards, K. L., Seitz, A. R., Lawrie, S. M., & Series, P. (2019). Acquisition of visual priors and induced hallucinations in chronic schizophrenia. Brain, 142(8), 2523-2537.

Karvelis, P., Seitz, A. R., Lawrie, S. M., & Series, P. (2018). Autistic traits, but not schizotypy, predict increased weighting of sensory information in Bayesian visual integration. eLife, 7, e34115.

[*: joint first authors]

A detailed publication list can be accessed here.

email: karvelis.povilas@gmail.com