Predictive Coding: A Unified Theory of Cognition in Health and Disease?

Introduction

The high incidence and burden of cognitive deficits worldwide has evoked substantial research into their molecular, cellular and behavioural manifestations. Their relationship, however, is often unclear. Yet, recent impetus has emerged from computational neuroscience, proposing complex frameworks linking microscale and macroscale normative brain function
and, consequently, proposing a re-characterisation of complex cognitive disorders from a novel perspective (Fig.1). One such framework is the predictive coding account of brain function. Hence, this essay will explore and critically evaluate theoretical advances and empirical evidence suggesting such a principled framework may allow the re-interpretation of healthy cognition and of seemingly unrelated clinical features of different cognitive disorders within a single theoretical perspective.

Figure 1: Understanding the physical substrate of cognitive disorders demands investigation across multiple levels of description with sufficient spatiotemporal resolution. The mechanistics of microscale function are most commonly studied using electrophysiological, optogenetic and biochemical approaches assessing single-neuron activity, intercellular/intracellular interactions, neuronal and non-neuronal populations, and cellular composition et cetera (see microscale), while the macroscale study draws from, inter alia, advances in neuroimaging to assess interactions between brain regions, extrinsic connectivity, and regional activity, while linking such observations to psychological phenomena and behavioural observations (see macroscale). For a more comprehensive understanding of brain function and how the physical, neurobiological substrate links to psychological operations, however, it is imperative to scrutinize how the dynamics on one level impact the activity and dynamics of the next. Hence, a better understanding of the mesoscale level of brain function is required to link micro and macroscale understanding of brain function. The mesoscale relates to neuronal circuitry with idiosyncratic patterns of synaptic connections enabling specific computations. While technological advances like optogenetics and dynamic causal modeling facilitate such investigation, computational neuroscience is outstandingly suited to investigate neural circuit dynamics and how these afford the computations for complex cognitive operations (see mesoscale). Illustration made with biorender.com.

A predictive coding account of brain function

While we find intellectual antecedents of predictive coding in Hegelian dialectics of “Unconscious Inference”[1] and Kant’s “Hyperpriors”[2], the idea was reinvigorated by the reformulation of the brain as a Bayesian inference-machine[3]. Homeostatic biological systems minimize “surprise” associated with their sensory states by continuously generating a mental model of the environment from which perceptual predictions arise that are compared to actual sensory information[4]. Any incongruencies between prediction and sensory input constitute prediction errors which are relayed to iteratively fine-tune such generative models and the dependencies and statistical regularities of the external world they reflect[4].

Substantial evidence indicates that the brain, indeed, appears to predict sensory input, as we see attenuated neuronal responses to predicted versus unpredicted stimuli in primates[5] and in humans through neuroimaging of sensory cortices in oddball-paradigms[6,7,8]. This, however, does not unambiguously validity predictive coding or discriminate it as most the plausible paradigm, as such tests do not exclude competing frameworks/models also postulating predictive abilities of the brain. Hence, it is sensible to assess if predictive coding, in particular, can be realised neurophysiologically: As the model’s functional logic must be enabled by an idiosyncratic neuroarchitecture and neurophysiology, we must consider empirical evidence relating the computational dependencies of predictive coding to neurophysiological dependencies through investigation of neurocircuitry.

Predictive coding predicts a hierarchical organisation of sensory cortices, with feedforward connections conveying prediction errors and feedback connections conveying predictions, suppressing prediction errors. This implies reciprocity and functional asymmetry of inter-layer connectivity. Backpropagated prediction errors update generative models to improve predictions, implying amenability and plasticity of generative models. Importantly, the “newsworthiness” of prediction errors, and thus their influence on generative models, is determined by their precision, with precise information being amplified, or attenuated in “noisy” and volatile environments, implying the need for neuromodulatory components[4,9].

There, indeed, is a satisfying correspondence between the model and neuronal connectivity (Fig.2). However, the attribution of computational roles to neuronal sub-populations is largely speculative, the computational roles of internal synaptic and axonal physiology and heterogenous dendritic integration are mostly neglected and numerous intrinsic and extrinsic connections were empirically identified in the brain that are not accounted for by the model[10]. Additionally, the neurobiology of precision-modulation remains unclear[11]. Lastly, current predictive coding models assume rate-code-dependent neuronal communication instead of spiking-code-dependent. These coding schemes, however, are heavily debated in the neuroscience community.

Hence, extending predictive coding to the spiking paradigm would require substantial reformulation. Hence, while predictive coding is a compelling unifying model of cognitive function, accounting for neuroanatomical, neurophysiological and neuropsychophysical attributes of the brain, such as classical/extra-classical receptive fields[12], end-stopping[4], bistable perception[13], motion-illusions[14], action-control[15], rhythmic perception[16], auditory processing[17] and even memory and learning[18], it lacks satisfactory empirical foundations to make such claims. Using local field-potentials, (pharmaco-)electroencephalography and, particularly, optogenetics and dynamic causal-modeling for causal investigation of neurocircuitry and neurophysiology, we may address such unclarities and better assess the plausibility of predictive coding accounts of brain functions in both health and disease.

Figure 2: Early studies of feline visual cortices revealed a processing hierarchy depicting reciprocal connections between constituent cortical areas[19]. This basic hierarchical principle is conserved, with minor alterations, in other sensory cortices. The simplified schematic of a cortical hierarchy (B) highlights hierarchical processing architectures, depicting expectation units, most likely deep pyramidal cells[10] and error units, most likely superficial pyramidal cells[10] with prediction errors being fed forward and predictions reflecting the current perceptual hypothesis being fed back down the cortical hierarchy. Prediction errors are contextualised through precision modulation by neuromodulatory cells (blue circle). Notably, the descending gain control of prediction errors rests on predictions of precision of prediction errors at lower levels, explaining why neuromodulatory cells are contacted by higher-level expectation units (yellow arrow). This connection is not shown in the schematic of the visual system processing hierarchy. In the visual system (A) and other sensory cortices, we, indeed, see morphological and functional asymmetry between forward and backward projections as neuroimaging[20] and reversible inactivation studies[21] confirmed the excitatory character of forward connections carrying sensory input from the retina to the visual cortex (black arrows), whereas backward connections, that “explain away” ascending signals, were often shown to be modulatory or inhibitory (magenta arrows). Note, that, for simplicity, higher levels of the visual processing hierarchy such as occipital and temporal lobe structures contacted by dorsal and ventral streams are not shown here. Associative plasticity is demonstrated empirically in this hierarchical processing system, which is an evident neurophysiological correlate to the updating of the generative models[18]. Lastly, we see prediction error-encoding, superficial pyramidal cells demonstrate various synaptic gain control mechanisms through dopamine receptors[22], NMDA receptors (red circles) as well as acetylcholine-based neuromodulation[11] by input of neuromodulatory cells from the ventral tegmental area and substantia nigra, accounting for the weighing of prediction errors by their expected precision in a context-sensitive fashion. Here, projections of the frontal eye fields to the oculomotor system through pontine nuclei in response to proprioceptive feedback have been indicated, allowing classical reflexes, highlighting the extension of predictive coding to active inference, where perception guides behavioural sequelae (actions) in order to resample sensory data to update the generative model and minimise future proprioceptive prediction errors. These simplified schematics can be nuanced by considering how expectation and error units can be deployed within canonical microcircuits of cortical columns present at every level of the cortical hierarchy, a neuroarchitectural theme which is replicated throughout the cortex[23] and which is also coherent with the circuitry implied by predictive coding ©. Functional and anatomical studies show input generally enters through granular layer 4, is relayed to layers 2/3 excitatory and inhibitory pyramidal cells encoding information corresponding to the updated generative model. These new predictions are conveyed through feedback connections back to infragranular pyramidal cells in layer 5/6 and cortical areas lower in the processing hierarchy. These supragranular cells also convey prediction errors to higher cortical areas through feed-forward connections. Neuromodulatory input is not depicted here, as the exact intrinsic circuitry for precision modulation is not fully established. Importantly, the segregation of forward and backward streams, which have different preferred oscillatory frequencies[24], highlighting their functional asymmetry[23], shows a remarkable correspondence with the circuitry implied by predictive coding. The green arrows indicate connections not accounted for by the model, such as deep-deep feedback pathways between adjacent layers of the cortical hierarchy[25]. Possibly, these may imply a separate prediction pathway allowing the downstream passage of predictions without modulation by supragranular prediction error units. The predictive model focuses on cortico-cortico connectivity, however, we also find many subcortico-cortico, particularly cortico-thalamic connections, which are not adequately integrated into predictive coding processing theories[26].

Predictive coding in memory — conflicting theories?

The hippocampus underlies episodic memory[27] while also continuously predicting sensory experience[28, 29], as seen in phase precession[30,31]. At the apex of hierarchical cortical processing, it appears to provide multimodal predictions based upon a generative model encompassing multisensory representations[18, 32]. These two cardinal functions, however, highlight a poorly scrutinized dichotomy: during episodic memory recall, the hippocampus is thought to excite the neocortex reinstating patterns of activity within neocortical networks[33,34,35], whereas predictive processing implies descending predictions inhibiting ascending input[4] which appears irreconcilable given the theorized unitary hippocampal code underlying both[36]. One way to solve this dialectic would be to hypothesize, that inhibition of neocortical cells through specific neocortical inhibitory interneurons would facilitate prediction whereas their disinhibition would reinstate neocortical activity and drive memory recall (Fig. 3). This could be tested electrophysiologically and biochemically (e.g.anterograde tracing) by assessing if excitatory hippocampal projections contact neocortical inhibitory interneurons (e.g.Somatostatin+-neurons) and whether disinhibitory mechanisms such as activation of vasoactive-intestinal-peptide+ interneurons and/or modulation of Parvalbumim+- interneurons are present in the neocortex.

Figure 3: Here, deep hippocampal pyramidal cell either target inhibitory somatostatin+-interneurons (SOM) allowing descending inhibition of neocortical superficial pyramidal cells required for prediction in predictive coding or vasoactive Intestinal Peptide+-interneurons (VIP) that disinhibit SOM+-interneurons or Parvalbumin+-interneurons (PV), hence, allowing the excitation of neocortical networks required for episodic memory recall.

Further concerning memory, mental schemas, frameworks of acquired knowledge stored in the neocortex that influence the assimilation of novel information[37], could also be regarded as generative models capturing regularities of the world. If we do so, active inference, which absorbs predictive coding into an action-perception-loop, describes a neurobiologically plausible way these schemas are used to assimilate evidence, learn affordances and assimilate evidence[0]. This speculation could be substantiated through refinements of the active inference model and a more sophisticated understanding of the neurocircuitry enabling mental schemas.

Predictive coding in disease

Given the substantial, yet not exhaustive, correspondence of networks and dependencies in the anatomical and theoretical domain of predictive coding, it is sensible to argue that disrupting specific structures should elicit a predictable, computational consequence and result in false inference allowing the interpretation of different cognitive deficits through a common framework.

The disconnection hypothesis has implicated predictive coding in schizophrenia, a cardinal characteristic of which is the production of false beliefs, more specifically, paranoid ideation, delusions, and hallucinations[38]. The hypothesis suggests that the key pathophysiology lies in abnormal modulation of NMDA-receptor-mediated synaptic efficacy by modulatory neurotransmitter systems[39], essentially characterising schizophrenia as a synaptic gain pathology[40]. This is supported empirically by genome-wide association studies showing that mutations in receptors of glutamate and dopamine, which are neuromodulatory[41,42] and aberrant NMDAR-subunit biosynthesis through serine-racemase mutations are genetic risk factors[43]. As superficial pyramidal cells, reporting prediction errors, contain these synaptic gain-control mechanisms (Fig.2) and the control of synaptic gain refers, computationally, to precision-encoding required for the selection of salient information updating and contextualizing our beliefs, it is sensible to argue that the aberrant precision modulation may compromise context-sensitive veridical inference, manifesting as delusional beliefs[44]. This is supported by behavioural and neuroimaging tests that assess predictive abilities (Fig.4).

This contrasts with the apparent precedence of internal generative models over sensory evidence as seen in physiological aging [45] and numerous pathologies demonstrating sensory attenuation, eliciting opposite effects in such tests (Fig. 4).

Figure 4: In schizophrenia, NMDA-R hypofunction and the proposed increase in dopamine activity[42,43] are evident correlates of reduced precision of representations of the prior (of predictions) and abnormally strong weighting of prediction error, referring to attributions of high reliability to predictive errors regardless of context (e.g. in ambiguous visual environments). Following the computationally logic of predictive coding models, this would hinder veridical inference through inappropriate predictive coding. This may have various behavioural manifestations such as a loss of mismatch negativity (B). Mismatch negativity refers to the electrophysiological activity in auditory event-related potentials in response to an “odd” sound, differing in frequency, loudness, or duration from the standard stimulus, after a repetitive sequence of sounds. In schizophrenia, this appears to be attenuated, meaning that the patients' response to a predicted and unpredicted stimulus is comparable[46] further reflecting the bias away from predictions. Force matching paradigms measure the strengths with which a bar is pressed when instructed to match the pressure from the other side. Control subjects overcompensate with force indicating functional and normative attenuation of sensory feedback through top-down predictions (A), whereas schizophrenics do not, consistent with the postulate that sensory consequences of their action are not attenuated by top-down predictions adequately[47]. Lastly, illusions rely on an agent’s internal model of the world and hence descending predictions, which is why schizophrenics, in which there is a bias away from prior beliefs, demonstrate resistance to illusions. In contrast, in circumstances where sensory input is attenuated, through metabolic hypofunction or cholinergic hypofunction in sensory areas (e.g. Lewy Body dementia), traumatic injuries (e.g. loss of limb) or damage to sensory structures (e.g. retinal damage in Charles Bonnet Syndrome), inflammation or physiological ageing-related changes, it is sensible to expect a bias towards prior beliefs and generative model-derived predictions, which affects the inferential process[45](C). Sensory attenuation partly uncouples cortical activity from sensory input and disrupts the finely-tuned balance between internal predictions comprising regularities of our environment captured over our lifetime and incoming sensory data impairing one’s predictive coding ability[48]. This is supported empirically by transfer entropy analysis showing an increased reliance on cross-modal generative models in aged populations, where we often see sensory attenuation, compared to younger individuals[45] and explains the increased susceptibility to illusions[45], reduced mismatch activity[49] and increasing overcompensation in force-matching with age[48]. In Lewy Body Dementia, we observe pareidolic hallucination in ambiguous environments.

With only about 6% of mild cognitive impairment cases diagnosed correctly, such tests like oddball-paradigms, multimodal illusions or force-matching may serve as early indices of cognitive decline in various pathologies, potentially facilitating future diagnoses and early interventions[47]. Hence, it would be valuable to sophisticate such tests, assessing different nuances of predictive coding in the light of cognitive disorders.

Lastly, the similarities in pathophysiology and behavioural symptoms between dementias imply that an overarching framework like predictive coding may find application. Indeed, we find that predictive abilities such as the micro-timescale anticipation of events are reduced in various dementias as shown in auditory mismatch paradigms[51]. This can be explained by, for example, breakdown of cholinergic projections in Alzheimer’s[52], thwarting the precision modulation of sensory evidence, more precisely, the gain of supragranular pyramidal cells and, thus, appropriate predictive coding[53]. This is also corroborated by pharmacological studies showing that cholinesterase-inhibitors can partially restore the mismatch response and other cognitive deficits associated with aberrant predictive coding in Alzheimer’s[53,54].

While these are legitimate attempts to connect pathophysiology and psychopathology, they rest on the very theoretical concept of predictive coding and future research should further our understanding of links between neuroanatomy, polygenic predisposition, synaptic gain control mechanisms, epigenetics, and how they relate to the parameters of predictive coding theory.

Future research

Often, the causative, neurobiological substrate of a cognitive deficit is unclear. Equally, we are in dire need of new drugs counteracting cognitive deficits. Here, the predictive coding framework may be invaluable in bridging these knowledge gaps through systematic integration of growing evidence of how predictive coding relates to its neuroanatomical substrates, and how drugs affect cognition and/or its physical substrate (Fig.5). This could be done in observational studies of patients treated with drugs targeting cognitive decline that undergo cognitive tests and post-mortem histopathological examinations or, interventionally, in preclinical models, although these rarely recapitulate all aspects of complex cognitive disorders.

Figure 5: (1) If we know how neuropathological changes relate to cognitive changes associated with aberrant predictive coding as can be measured using the tests listed above (e.g. cholinergic neuron degeneration reduces precision modulation of prediction errors), this could give insight as to which components of the neural system to target pharmacologically (e.g. Acetylcholine insufficiency through cholinesterase inhibitors), hence contributing to drug development [pink arrow]. (2) If we know how drugs, given we are aware of their mechanism of action, affect cognitive parameters related to predictive coding, this could elucidate the neuropathological substrate of these deficits, given the congruence between networks and dependencies in the anatomical and theoretical domain of predictive coding that has been established above. If we, for example, see that glycine modulatory site agonists, that we know induce NMDAR-activation, improve cognitive symptoms such as hallucinations which are explainable by predictive coding, we may infer that NMDA-receptors play a crucial role in Schizophrenia pathogenesis, hence contributing to the mechanistic investigation of the physical substrate of cognition and cognitive deficits [blue arrow]. (3) Lastly, if we do not know yet exactly how a component of a neural system underlying a particular cognitive ability associated with predictive coding is related to computational components in the theoretical domain of predictive coding, like the role of deep-deep feedback connections between adjacent layers of the cortical hierarchy (see figure 2), we may target this neural system component pharmacologically (or through lesions, gene manipulations, etc), and assess which aspect of predictive coding is affected (e.g. increased reliance on priors, reduced precision of prediction errors et cetera), allowing us to substantiate the relationship between this neural component and a computational role implied by predictive coding, or even to refine the predictive coding model entirely [green arrow].

Conclusion

This essay demonstrated that diverse neurocognitive deficits could be interpreted through a common framework by appreciating that their emergence may rest on the disruption of fine-tuned cortical circuitry, enabling inferential computation, through neurochemical imbalances, widespread atrophy, or changes in connectivity.

Technological advancements in the study of microcircuitry and theoretical progress regarding predictive coding may, in the future, drive a two-way exchange where a better understanding of microcircuitry enhances theoretical modeling and predictive coding theory may provide novel impetus for studies of neural circuitry and, together, prompt novel investigations yielding valuable clinical insights.

References

[0] Conversations with Dr Karl Friston (email).

[1] Eldridge, R. (2014). Hegel’s Account of the Unconscious and Why It Matters. The Review of Metaphysics, Vol. 3, pp. 491–515.

[2] Swanson, L. (2016). The Predictive Processing Paradigm Has Roots in Kant. Frontiers in Systems Neuroscience, 10, pp 445–466.

[3] Friston, K. (2012.) The history of the future of the Bayesian brain. NeuroImage, 62(2), pp.1230- 1233.

[4] Friston, K. (2005). A theory of cortical responses. Philosophical Transactions of the Royal Society B: Biological Sciences, 360(1456), pp.815–836.

[5] Meyer, T. and Olson, C. (2011). Statistical learning of visual transitions in monkey inferotemporal cortex. Proceedings of the National Academy of Sciences, 108(48), pp.19401- 19406.

[6] Alink, A., Schwiedrzik, C., Kohler, A., Singer, W. and Muckli, L. (2010). Stimulus Predictability Reduces Responses in Primary Visual Cortex. Journal of Neuroscience, 30(8), pp.2960–2966.

[7] den Ouden, H., Daunizeau, J., Roiser, J., Friston, K. and Stephan, K. (2010). Striatal Prediction Error Modulates Cortical Coupling. Journal of Neuroscience, 30(9), pp.3210–3219.

[8] Todorovic, A. and de Lange, F. (2012). Repetition Suppression and Expectation Suppression Are Dissociable in Time in Early Auditory Evoked Fields. Journal of Neuroscience, 32(39), pp.13389- 13395.

[9] Parr, T., Friston, K.J. (2019). Generalised free energy and active inference. Biol Cybern 113, pp.495– 513.

[10] Shipp, S. (2016). Neural Elements for Predictive Coding. Frontiers in Psychology, 7.

[11] Moran, R., Campo, P., Symmonds, M., Stephan, K., Dolan, R. and Friston, K. (2013). Free Energy, Precision and Learning: The Role of Cholinergic Neuromodulation. Journal of Neuroscience, 33(19), pp.8227–8236.

[12] Rao, R. and Ballard, D. (1999). Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience, 2(1), pp.79–87.

[13] Weilnhammer, V., Stuke, H., Hesselmann, G., Sterzer, P. and Schmack, K. (2017). A predictive coding account of bistable perception — a model-based fMRI study. PLOS Computational Biology, 13(5), p.e1005536.

[14] van Heusden, E., Harris, A., Garrido, M. and Hogendoorn, H. (2019). Predictive coding of visual motion in both monocular and binocular human visual processing. Journal of Vision, 19(1), p.3.

[15] Kahl, S. and Kopp, S. (2018). A Predictive Processing Model of Perception and Action for Self- Other Distinction. Frontiers in Psychology, 9.

[16] Vuust, P. and Witek, M. (2014). Rhythmic complexity and predictive coding: a novel approach to modeling rhythm and meter perception in music. Frontiers in Psychology, 5.

[17] Denham, S. and Winkler, I. (2020). Predictive coding in auditory perception: challenges and unresolved questions. European Journal of Neuroscience, 51(5), pp.1151–1160.

[18] Barron, H., Auksztulewicz, R. and Friston, K. (2020). Prediction and memory: A predictive coding account. Progress in Neurobiology, 192, p.101821.

[19] Maunsell, J. and van Essen, D. (1983). The connections of the middle temporal visual area (MT) and their relationship to a cortical hierarchy in the macaque monkey. The Journal of Neuroscience, 3(12), pp.2563–2586.

[20] Buchel, C. (1997). Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI. Cerebral Cortex, 7(8), pp.768–778.

[21] Girard, P. and Bullier, J. (1989). Visual activity in area V2 during reversible inactivation of area 17 in the macaque monkey. Journal of Neurophysiology, 62(6), pp.1287–1302.

[22] Lidow, M., Goldman-Rakic, P., Gallager, D. and Rakic, P. (1991). Distribution of dopaminergic receptors in the primate cerebral cortex: Quantitative autoradiographic analysis using [3H]raclopride, [3H]spiperone and [3H]SCH23390. Neuroscience, 40(3), pp.657–671.

[23] Bastos, A., Usrey, W., Adams, R., Mangun, G., Fries, P. and Friston, K. (2012). Canonical Microcircuits for Predictive Coding. Neuron, 76(4), pp.695–711.

[24] Arnal, L. and Giraud, A., 2021. Cortical Oscillations and Sensory Predictions. Cell. [25] da Costa, N. (2010). Whose cortical column would that be?. Frontiers in Neuroanatomy,.

[26] Markov, N., Vezoli, J., Chameau, P., Falchier, A., Quilodran, R., Huissoud, C., Lamy, C., Misery, P., Giroud, P., Ullman, S., Barone, P., Dehay, C., Knoblauch, K. and Kennedy, H. (2013). Anatomy of hierarchy: Feedforward and feedback pathways in macaque visual cortex. Journal of Comparative Neurology, 522(1), pp.225–259.

[27] Squire, L. and Zola-Morgan, S. (1991). The Medial Temporal Lobe Memory System. Science, 253(5026), pp.1380–1386.

[28] Stachenfeld, K., Botvinick, M. and Gershman, S. (2017). The hippocampus as a predictive map. Nature Neuroscience, 20(11), pp.1643–1653.

[29] Pezzulo, G., Kemere, C. and van der Meer, M. (2017). Internally generated hippocampal sequences as a vantage point to probe future-oriented cognition. Annals of the New York Academy of Sciences, 1396(1), pp.144–165.

[30] Jensen, O. and Lisman, J. (1996). Hippocampal CA3 region predicts memory sequences: accounting for the phase precession of place cells. Learning & Memory, 3(2–3), pp.279–287.

[31] Tsodyks, M., Skaggs, W., Sejnowski, T. and McNaughton, B. (1998). Population dynamics and theta rhythm phase precession of hippocampal place cell firing: A spiking neuron model. Hippocampus, 6(3), pp.271–280.

[32] Barron, H., Auksztulewicz, R. and Friston, K. (2020). Prediction and memory: A predictive coding account. Progress in Neurobiology, 192, p.101821.

[33] Moscovitch, M., Cabeza, R., Winocur, G. and Nadel, L. (2016). Episodic Memory and Beyond: The Hippocampus and Neocortex in Transformation. Annual Review of Psychology, 67(1), pp.105–134.

[34] Rugg, M. and Vilberg, K. (2013). Brain networks underlying episodic memory retrieval. Current Opinion in Neurobiology, 23(2), pp.255–260.

[35] Teyler, T. and Rudy, J. (2007). The hippocampal indexing theory and episodic memory:Updating the index. Hippocampus, 17(12), pp.1158–1169.

[36] Barron, H., Auksztulewicz, R. and Friston, K. (2020). Prediction and memory: A predictive coding account. Progress in Neurobiology, 192, p.101821.

[37] Tse, D., Langston, R., Kakeyama, M., Bethus, I., Spooner, P., Wood, E., Witter, M. and Morris, R. (2007). Schemas and Memory Consolidation. Science, 316(5821), pp.76–82.

[38] Fletcher, P. and Frith, C. (2008). Perceiving is believing: a Bayesian approach to explaining the positive symptoms of schizophrenia. Nature Reviews Neuroscience, 10(1), pp.48–58.

[39] Friston, K., Brown, H., Siemerkus, J. and Stephan, K. (2016). The dysconnection hypothesis. Schizophrenia Research, 176(2–3), pp.83–94.

[40] Adams, R., Pinotsis, D., Tsirlis, K., Unruh, L., Mahajan, A., Horas, A., Convertino, L., Summerfelt, A., Sampath, H., Du, X., Kochunov, P., Ji, J., Repovs, G., Murray, J., Friston, K., Hong, L. and Anticevic, A. (2021). Computational Modeling of Electroencephalography and Functional Magnetic Resonance Imaging Paradigms Indicates a Consistent Loss of Pyramidal Cell Synaptic Gain in Schizophrenia. Biological Psychiatry,.

[41] Rubio, M., Drummond, J. and Meador-Woodruff, J. (2012). Glutamate Receptor Abnormalities in Schizophrenia: Implications for Innovative Treatments. Biomolecules and Therapeutics, 20(1), pp.1–18.

[42] Seeman, P. and Kapur, S. (2000). Schizophrenia: More dopamine, more D2 receptors. Proceedings of the National Academy of Sciences, 97(14), pp.7673–7675.

[43] Labrie, V., Fukumura, R., Rastogi, A., Fick, L., Wang, W., Boutros, P., Kennedy, J., Semeralul, M., Lee, F., Baker, G., Belsham, D., Barger, S., Gondo, Y., Wong, A. and Roder, J. (2009). Serine racemase is associated with schizophrenia susceptibility in humans and in a mouse model. Human Molecular Genetics, 18(17), pp.3227–3243.

[44] Corlett, P. and Fletcher, P. (2014). Computational psychiatry: a Rosetta Stone linking the brain to mental illness. The Lancet Psychiatry, 1(5), pp.399–402.

[45] Chan, J., Wibral, M., Stawowsky, C., Brandl, M., Helbling, S., Naumer, M., Kaiser, J. and Wollstadt, P. (2021). Predictive Coding Over the Lifespan: Increased Reliance on Perceptual Priors in Older Adults — A Magnetoencephalography and Dynamic Causal Modeling Study. Frontiers in Aging Neuroscience, 13.

[46] Pezzulo, G., Kemere, C. and van der Meer, M. (2017). Internally generated hippocampal sequences as a vantage point to probe future-oriented cognition. Annals of the New York Academy of Sciences, 1396(1), pp.144–165.

[47] Shergill, S., Samson, G., Bays, P., Frith, C. and Wolpert, D. (2005). Evidence for Sensory Prediction Deficits in Schizophrenia. American Journal of Psychiatry, 162(12), pp.2384–2386.

[48] Wolpe, N., Ingram, J., Tsvetanov, K., Geerligs, L., Kievit, R., Henson, R., Wolpert, D. and Rowe, J. (2016). Ageing increases reliance on sensorimotor prediction through structural and functional differences in frontostriatal circuits. Nature Communications, 7(1).

[49] Cheng, C., Hsu, W. and Lin, Y. (2013). Effects of physiological aging on mismatch negativity: A meta-analysis. International Journal of Psychophysiology, 90(2), pp.165–171.

[50] Uchiyama, M., Nishio, Y., Yokoi, K., Hirayama, K., Imamura, T., Shimomura, T. and Mori, E. (2012). Pareidolias: complex visual illusions in dementia with Lewy bodies. Brain, 135(8), pp.2458–2469.

[51] Ruzzoli, M., Pirulli, C., Mazza, V., Miniussi, C. and Brignani, D. (2016). The mismatch negativity as an index of cognitive decline for the early detection of Alzheimer’s disease. Scientific Reports, 6(1).

[52] Mufson, E., Counts, S., Perez, S. and Ginsberg, S. (2008). Cholinergic system during the progression of Alzheimer’s disease: therapeutic implications. Expert Review of Neurotherapeutics, 8(11), pp.1703–1718.

[53] Moran, R., Campo, P., Symmonds, M., Stephan, K., Dolan, R. and Friston, K. (2013). Free Energy, Precision and Learning: The Role of Cholinergic Neuromodulation. Journal of Neuroscience, 33(19), pp.8227–8236.

[54] Riekkinen Jr., P., Pääkkönen, A., Karhu, J., Partanen, J., Soininen, H., Laakso, M., Riekkinen Sr., P. and Riekkinen Sr., P. (1997). THA disrupts mismatch negativity in Alzheimer disease. Psychopharmacology, 133(2), pp.203–206.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Fynn Comerford

Neuroscience at The University of Edinburgh | Founder of edventure | iGEM 2020 | Videographer