Computational Psychiatry Course - 29th-30th April 2015 Venue: Basement Lecture Theatre, 33 Queen Square, London, WC1N 3BG The computational anatomy of psychosis Karl Friston Abstract We will use schizophrenia as a case study of computational psychiatry. We first review the basic phenomenology and pathophysiological theories of schizophrenia. These motivate the choice of a formal or computational framework within which to understand the symptoms and signs of schizophrenia. This framework is the Bayesian brain or Bayesian decision theory. We will focus on the encoding of uncertainty or precision within predictive coding implementations of the Bayesian brain to demonstrate how computational approaches can disclose the nature of hallucinations and delusions. The symptoms and signs of schizophrenia Bleuler Dysmorphophobia Delusions Delusional mood False beliefs Delusional systems Depersonalisation Psychomotor property Cognitive deficits Compulsions Intrusive thoughts Obsessional beliefs Hallucinations False percepts Affective symptoms Dissociation syndromes Thought disorder Capgras syndrome Listening of associations Disintegration of the psyche Functional medical syndromes Anxiety Persecutory beliefs … Aberrant beliefs and false inference Soft neurological signs Abnormal eye movements Abnormal mismatch negativity Pathophysiological and aetiological theories Bleuler Genetic Dopamine hypothesis Abnormal plasticity Aberrant salience Glutamate hypothesis NMDA receptor dysfunction Aberrant synchrony GABAergic hypothesis Aberrant gain control Abnormal E-I balance Neurodevelopmental Aberrant neuromodulation and synaptic gain control Psychotomimetic drugs Psychosocial Autoimmune Dysconnection in schizophrenia: from abnormal synaptic plasticity to failures of self-monitoring. Schizophr Bull. 2009 May;35(3):509-27 Klaas E. Stephan, Karl J. Friston and Chris D. Frith Wernicke C. Grundrisse der Psychiatrie. 1906: Sejunction – disruption of associative connectivity Anatomical disconnection Bleuler E. Dementia Praecox oder Gruppe der Schizophrenien, 1911: Disintegration – of conscious processing (the psyche) Functional dysconnection Aberrant neuromodulation and synaptic gain control Which computational (formal) framework? Reinforcement learning, optimal control and expected utility theory Information theory and minimum redundancy Pavlov Barlow Self-organisation, synergetics and allostasis Haken Bayesian brain, Bayesian decision theory and predictive coding Helmholtz Which computational (formal) framework? Reinforcement learning, optimal control and expected utility theory Information theory and minimum redundancy Pavlov Barlow Self-organisation, synergetics and allostasis Haken Bayesian brain, Bayesian decision theory and predictive coding Helmholtz Active inference, predictive coding and precision Precision and false inference Simulations of : Auditory perception (and omission related responses) Smooth pursuit eye movements (under occlusion) Sensory attenuation (and the force matching illusion) “Objects are always imagined as being present in the field of vision as would have to be there in order to produce the same impression on the nervous mechanism” - von Helmholtz Hermann von Helmholtz Richard Gregory Geoffrey Hinton From the Helmholtz machine to the Bayesian brain Thomas Bayes Richard Feynman Bayesian filtering and predictive coding t D prediction update s g ( ) prediction error Making our own sensations sensations – predictions Prediction error Action Perception Changing sensations Changing predictions Generative models Discrete states Continuous states v(3) , v(2) x(2) x(2) Action (2) v a0 , Control states , at 1 ut , (1) v C , uT B x(1) x(1) v(1) v(0) Hidden states (i ) ot 1 ot st 1 Variational Bayes (belief updating) D (i ) st A Bayesian filtering (predictive coding) (i ) st 1 (i ) (i ) (i ) ( i ) 0 st st (ln A ot ln B(at 1 )st 1 ) Hierarchical generative models what (3) v(3) v Ascending prediction errors the (2) v (2) x(2) x x(2) (2) v(2) v v(1) (1) x(1) x x(1) (1) v(1) v v(0) where Descending predictions D Sensory fluctuations David Mumford Predictive coding with reflexes a a s v(1) oculomotor signals reflex arc Action proprioceptive input pons retinal input Perception Prediction error (superficial pyramidal cells) frontal eye fields Top-down or backward predictions Precision geniculate (i ) VTA v(i ) (vi ) v(i ) (vi ) (v(i 1) g (i ) ( x(i ) , v(i ) )) x(i ) (xi ) x(i ) (xi ) (D x(i ) f (i ) ( x(i ) , v(i ) )) Conditional predictions (deep pyramidal cells) Bottom-up or forward prediction error visual cortex (i ) v(i ) Dv(i ) v (i ) (i ) v(i 1) x(i ) D x(i ) x (i ) (i ) Unconditioned stimulus (US) Bayesian belief updating a Motor Cortex Condition stimulus (CS) ot Striatum st Simulated (US) responses 400 350 Inferotemporal Cortex Q 300 Rate 250 200 150 Prefrontal Cortex 100 50 0 VTA/SN 1 2 3 4 Peristimulus time (sec) Simulated (CS & US) responses 400 Action selection Incentive salience st (ln A ot ln B(at 1 ) st 1 ) ( Q) Q 350 300 250 Rate Perception 200 150 100 50 0 1 2 3 Peristimulus time (sec) 4 Active inference, predictive coding and precision Precision and false inference Simulations of : Auditory perception (and omission related responses) Smooth pursuit eye movements (under occlusion) Sensory attenuation (and the force matching illusion) De-compensation (trait abnormalities) + Neuromodulatory failure (of sensory attenuation) Attenuated violation responses Loss of perceptual Gestalt SPEM abnormalities Psychomotor poverty Resistance to illusions Hallucinations Delusions - Compensation (to psychotic state) Active inference, predictive coding and precision Precision and false inference Simulations of : Auditory perception (and omission related responses) Smooth pursuit eye movements (under occlusion) Sensory attenuation (and the force matching illusion) Neuronal hierarchy Generative model (1) 1 v (1) 2 v f (2) 18 x2(2) 18 x1(2) (2) (2) (2) (2) 32 x1 2 x3 x1 x2 2 x (2) x (2) 8 x(2) 3 3 1 2 f (1) 18 x2(1) 18 x1(1) (1) (1) (1) (1) (1) v1 x1 2 x3 x1 x2 2 x (1) x (1) v (1) x (1) 2 3 1 2 x (1) s g (1) 2(1) 1 x3 s2 x (2) v (1) g (2) 2(2) 1(1) x3 v2 Syrinx Predictive coding v( i ) Dv( i ) v ( i ) ( i ) v( i 1) x( i ) D x( i ) x (i ) ( i ) sonogram percept v(0) v(1) prediction error micro-volts) 8 LFP ( Frequency (KHz) Frequency (Hz) 10 6 4 2 0 -2 -4 -6 0.5 1 Time (sec) 1.5 500 1000 1500 peristimulus time (ms) 2000 percept response to violation Omission related responses, MMN and hallucinosis 100 5000 4500 LFP (micro-volts) Frequency (Hz) 50 4000 3500 3000 0 -50 2500 2000 0.5 1 time (sec) -100 1.5 percept 500 1000 1500 peristimulus time (ms) 2000 attenuated mismatch negativity 100 5000 4500 Reduced prior precision LFP (micro-volts) Frequency (Hz) 50 4000 3500 3000 0 -50 2500 2000 0.5 1 time (sec) -100 1.5 500 percept 1000 1500 peristimulus time (ms) 2000 hallucination 100 5000 4500 Compensatory attenuation of sensory precision LFP (micro-volts) Frequency (Hz) 50 4000 3500 3000 0 -50 2500 2000 0.5 1 time (sec) 1.5 -100 500 1000 1500 peristimulus time (ms) 2000 Active inference, predictive coding and precision Precision and false inference Simulations of : Auditory perception (and omission related responses) Smooth pursuit eye movements (under occlusion) Sensory attenuation (and the force matching illusion) Smooth pursuit eye movements Angular direction of target x t(1) Angular direction of gaze in extrinsic coordinates x (1) o Angular position of target in intrinsic xt(1) x (1) o coordinates so x(1) o proprioceptive input pons retinal input st visual channels time ù és ù é xo(1) ú+ wv(1) s = ê o ú= ê êst ú êO( xt(1) ) ×exp(- ([- 8,¼ ,8] + xo(1) - xt(1) ) 2 )ú ë û ë û (1) ù (1) éx& é ù xo¢ ê o ú ê ú (1) (1) (1) ê1 (v - x (1) ) - 1 x ¢(1) ú+ w(1) ¢ ú x& = êêx& = o o o x 4 2 ú ê ú (1) (1) êx&(1) ú ê ú v - xt ú ëê ú ëê t û û és ù s = ê o ú= êst ú ë û éx&o ù ê ú ú= x&= êx&¢ ê oú êx&ú ë tû é ù xo ê ú+ ω (1) v 2 êO(xt ) ×exp(- ([- 8, K ,8] + x o - xt ) )ú ë û é x¢ ù o ê ú ê1 a - 1 x o¢ú+ ω (1) x 4 8 ê ú Generative process ê v- x ú ú t û ëê v (1) = x1(2) + wv(2) éx&(2) ù x&(2) = ê 1(2) ú= êx& ú ë2 û v (2) = h + wv(3) 1 8 é x (2) ù v (2) ê 2 (2) ú+ wx(2) ê- x ú ë 1 û Generative model Eye movements under occlusion and reduced prior precision Angular position displacement (degrees) 2 1 target eye 0 -1 eye (reduced precision) -2 500 1000 1500 2000 2500 3000 2000 2500 3000 velocity (degrees per second) Angular velocity 50 40 30 20 10 0 -10 -20 500 1000 1500 time (ms) displacement (degrees) Paradoxical responses to violations target and oculomotor angles 2 eye 1 target 0 eye (reduced precision) -1 -2 100 200 300 400 500 600 time (ms) 700 800 900 1000 900 1000 velocity (degrees per second) target and oculomotor velocities 30 20 10 0 -10 -20 -30 100 200 300 400 500 600 time (ms) 700 800 Active inference, predictive coding and precision Precision and false inference Simulations of : Auditory perception (and omission related responses) Smooth pursuit eye movements (under occlusion) Sensory attenuation (and the force matching illusion) Making your own sensations s x s p i s ss xi xe xi vi 14 xi x x 1 xe ve 4 xe v v i v ve a s xi s p ωs ss x i v e x xi (a ) 14 xi ω x sp ω s ~ N (0, e 8 I ) ω x ~ N (0, e 8 I ) ss s ~ N (0, e I ) x ~ N (0, e4 I ) xi Generative process ve 8 ( xi vi ) v ~ N (0, e6 I ) Generative model x v sensorimotor cortex descending predictions mx descending sensory predictions descending modulation thalamus prefrontal cortex mv ascending prediction errors v descending motor predictions s v a sp ss motor reflex arc Self-made acts prediction and error hidden states 2 xi 2 1.5 ss sp 1 1.5 1 0.5 0.5 0 0 -0.5 -0.5 5 10 15 20 25 30 xe 5 10 Time (bins) 15 20 25 30 Time (bins) Sensory attenuation hidden causes 1 perturbation and action 1 vi a 0.8 0.6 0.5 ve 0.4 0.2 0 0 -0.2 -0.4 -0.5 -0.6 -0.8 5 10 15 20 Time (bins) 25 30 5 10 15 20 Time (bins) 25 30 and psychomotor poverty prediction and error hidden states 2 2 1.5 1.5 1 1 0.5 0.5 0 0 -0.5 -0.5 5 10 15 20 25 30 5 10 time 15 20 25 30 time Failure of sensory attenuation hidden causes perturbation and action 1 1 0.8 0.6 0.5 0.4 0.2 0 0 -0.2 -0.4 -0.5 -0.6 -0.8 5 10 15 time 20 25 30 5 10 15 time 20 25 30 Perceived as less prediction and error 2 hidden states 2 1.5 Reproduced as more hidden states prediction and error 2 2 1.5 1.5 1 1 0.5 0.5 0 0 -0.5 -0.5 1.5 1 1 0.5 0.5 0 0 -0.5 -0.5 10 20 30 40 50 60 10 20 Time (bins) 30 40 50 60 10 20 Time (bins) 30 40 50 60 10 20 Time (bins) Sensory attenuation 30 40 50 60 Time (bins) Force matching illusion Intrinsic and extrinsic hidden causes 2 2 perturbation and action hidden causes perturbation and action 1.5 1.5 1 1 0.5 0.5 1.5 1.5 1 1 0.5 0.5 0 0 0 0 -0.5 -0.5 -0.5 10 20 30 40 Time (bins) 50 60 -0.5 10 20 30 40 Time (bins) 50 60 10 20 30 40 Time (bins) 50 60 10 20 30 40 Time (bins) 50 60 Compensated failures of sensory attenuation Normal subjects 3 Simulated Empirical (Shergill et al) Self-generated(matched) force Self-generated(matched) force 2.5 2 1.5 1 Schizophrenic subjects 0.5 0 0 0.5 1 1.5 2 External (target) force 2.5 3 External (target) force prediction and error 3.5 3 3 2.5 2.5 2 2 1.5 1.5 1 1 0.5 0.5 0 0 -0.5 10 20 30 40 50 hidden states 3.5 60 -0.5 10 20 Time (bins) 30 40 50 60 Time (bins) Failure of sensory attenuation and delusions of control hidden causes 3.5 3.5 3 3 2.5 2.5 2 2 1.5 1.5 1 1 0.5 0 0.5 -0.5 0 -1 perturbation and action 10 20 30 40 Time (bins) 50 60 -0.5 10 20 30 40 Time (bins) 50 60 A computational account of delusions of agency • We act by predicting our action to create (attenuated) prediction errors that are suppressed reflexively • A failure of sensory attenuation subverts our predictions and precludes action (psychomotor poverty) • Compensatory increases in prior precision reinstate (unattenuated) prediction errors • Unattenuated prediction errors can only be explained by (antagonistic) external forces (delusions of control and made acts) Summary Signs (of trait abnormalities) + - Neuromodulatory failure (of sensory attenuation) Attenuated violation responses Loss of perceptual Gestalt SPEM abnormalities Psychomotor poverty Resistance to illusions Symptoms (of psychotic state) Hallucinations Delusions Summary • What is the functional deficit? False inference due to aberrant encoding of precision A neuromodulatory failure of postsynaptic excitability: • What is the pathophysiology? • • • • • How can we measure it? Modelling of behaviour and noninvasive brain responses • What is the aetiology? • What is the intervention? • • • • Aberrant DA/NMDA subunit interactions Aberrant synchronous gain and fast (gamma) dynamics Aberrant cortical gain control and E-I (GABAergic) balance Aberrant dendritic integration (neuro-morphology) Computational modelling of choice behaviour Computational fMRI Dynamic casual modelling of intrinsic (precision) gain control … Thank you And thanks to collaborators: And colleagues: Rick Adams Ryszard Auksztulewicz Andre Bastos Sven Bestmann Harriet Brown Jean Daunizeau Mark Edwards Chris Frith Thomas FitzGerald Xiaosi Gu Stefan Kiebel James Kilner Christoph Mathys Jérémie Mattout Rosalyn Moran Dimitri Ognibene Sasha Ondobaka Will Penny Giovanni Pezzulo Lisa Quattrocki Knight Francesco Rigoli Klaas Stephan Philipp Schwartenbeck Micah Allen Felix Blankenburg Andy Clark Peter Dayan Ray Dolan Allan Hobson Paul Fletcher Pascal Fries Geoffrey Hinton James Hopkins Jakob Hohwy Mateus Joffily Henry Kennedy Simon McGregor Read Montague Tobias Nolte Anil Seth Mark Solms Paul Verschure And many others Prefrontal input control subjects - predictable control subjects - unpredictable schizophrenia - predictable schizophrenia - unpredictable PC PC IT IT Effects of predictability on recurrent inhibition 1.5 V5 control subjects schizophrenics 1 V5 V1 log modulation 0.5 0 -0.5 -1 -1.5 -2 Visual input V1 R V5 L V5 R IT cortical source L IT R PC L PC