Introducing Algorithmic Circuit Psychiatry
A Framework for Linking Mind, Brain, and Treatment
Three years ago, I watched a patient with treatment-resistant schizophrenia experience their first significant symptom improvement in over a decade. The breakthrough came not from a new antipsychotic, but from Cobenfy—a muscarinic M1/M4 receptor agonist that targets the very thalamocortical circuits my lab has spent years mapping. As her delusions began to fade and her cognitive function improved, I realized we were witnessing something profound: the convergence of circuit-level understanding with algorithmic insight, translated into therapeutic action.
This moment crystallized a conviction I've been developing throughout my career studying brain circuits, from my early work on sleep and memory consolidation to my current focus on thalamocortical dysfunction in psychiatric illness. We need a new framework for psychiatry—one that bridges the gap between brain circuits and the algorithms they implement, connecting both to meaningful clinical intervention.
I call this Algorithmic Circuit Psychiatry.
Why Now? Why a New Framework?
Our field stands at a crossroads. The era of "chemical imbalance" models is behind us, but what comes next remains fragmented across two powerful yet incomplete approaches:
Computational psychiatry has revolutionized how we think about behavior, giving us sophisticated models of decision-making, learning, and belief formation. These models can predict symptoms, guide treatment selection, and even forecast treatment response. But too often, they remain divorced from the neural hardware that implements them—elegant mathematical descriptions floating above biological reality.
Circuit psychiatry has delivered unprecedented precision in mapping brain pathways, identifying dysfunction patterns, and developing targeted neuromodulation approaches. We can now manipulate specific neural circuits with remarkable precision. Yet circuit diagrams alone rarely tell us what computations are going awry, or how disrupted connectivity patterns give rise to the complex phenomenology of psychiatric illness.
What we desperately need is a framework that bridges these levels—a way to understand how specific brain circuits implement the algorithms that generate thought, belief, emotion, and behavior, and how dysfunction in these implementations leads to the symptoms we observe in the clinic.
This is the essence of Algorithmic Circuit Psychiatry: understanding the brain as an information-processing system where specific circuit architectures implement identifiable algorithms, and where psychiatric symptoms emerge from algorithmic failures implemented in dysfunctional circuits.
Circuits as Algorithmic Substrates: Lessons from the Thalamus
My journey toward this framework began with a seemingly simple question: what does the thalamus actually do? For decades, it was dismissed as a mere "relay station"—a passive conduit for sensory information flowing to cortex. But as my lab began systematically probing thalamocortical circuits, a different picture emerged.
The thalamus, we discovered, is an active computational hub that implements sophisticated algorithms for attention, working memory, and cognitive control. Our work has shown that thalamic circuits don't just relay information—they dynamically filter, integrate, and route information based on behavioral context and internal state. When we reversibly inactivate specific thalamic nuclei during cognitive tasks, we don't see simple sensory deficits. Instead, we observe precise disruptions in attention, working memory, and cognitive flexibility—the very functions that are impaired in psychiatric illness.
This led to a crucial insight: thalamic circuits implement key algorithms for cognitive control. The mediodorsal thalamus, for instance, doesn't just connect prefrontal cortex to other brain regions—it actively participates in maintaining and updating working memory representations, implementing something akin to a dynamic gating mechanism that controls information flow based on task demands.
But the real breakthrough came when we began linking these circuit-level insights to psychiatric symptoms. In schizophrenia, we consistently observe thalamocortical dysfunction—reduced thalamic volume, altered connectivity, disrupted oscillations. Using our algorithmic framework, we could now ask: what if schizophrenia involves dysfunction in the specific algorithms that thalamocortical circuits implement?
From Thalamocortical Algorithms to Psychiatric Symptoms
Consider delusions—one of the most challenging symptoms to understand mechanistically. Traditional approaches focus on neurotransmitter imbalances or connectivity disruptions. But what if we think algorithmically?
Delusions can be understood as failures in Bayesian belief updating—the fundamental algorithm by which we integrate new evidence with prior beliefs to form updated beliefs about the world. In healthy brains, this process is implemented through dynamic interactions between thalamic and cortical circuits. The thalamus continuously samples sensory and internal information, while cortical circuits maintain hierarchical models of the world. When new evidence arrives, thalamocortical loops implement a sophisticated updating algorithm that weighs evidence against priors, updates beliefs accordingly, and maintains uncertainty estimates.
In schizophrenia, our research suggests this algorithm becomes pathologically biased. Dysfunctional thalamocortical circuits may implement aberrant belief updating, where weak evidence is given excessive weight (leading to delusions) or where prior beliefs become too rigid (leading to cognitive inflexibility). The recent success of muscarinic M1/M4 agonists like Cobenfy makes perfect sense in this framework—these drugs specifically target the cholinergic modulation of thalamocortical circuits, potentially restoring more adaptive belief updating algorithms.
This isn't just theoretical speculation. My lab has been systematically testing these ideas using a combination of approaches:
Optogenetic manipulation of specific thalamocortical pathways during cognitive tasks to causally link circuit function to algorithmic performance
Computational modeling that captures how thalamic dynamics implement belief updating and working memory algorithms
Translational studies examining how thalamocortical dysfunction in psychiatric illness maps onto specific algorithmic impairments
The results consistently support the same conclusion: psychiatric symptoms reflect algorithmic failures implemented in dysfunctional circuits.
Beyond the Thalamus: A Broader Algorithmic Framework
While my own work has focused heavily on thalamocortical circuits, the Algorithmic Circuit Psychiatry framework extends far beyond any single brain region. Across the brain, we can identify circuit motifs that implement core algorithmic functions:
Striatal Reinforcement Learning Circuits The basal ganglia implement sophisticated reinforcement learning algorithms—updating value estimates, computing prediction errors, and guiding action selection based on expected outcomes. In addiction and OCD, these circuits become pathologically biased, implementing algorithms where certain actions or stimuli acquire excessive value or where habit formation overrides goal-directed behavior. Our understanding of these circuits as algorithmic learners with altered plasticity rules directly guides both pharmacological interventions (targeting dopaminergic modulation of learning) and behavioral interventions (exposure therapy as algorithm retraining).
Hippocampal-Cortical Generative Models The hippocampus, in conjunction with cortical areas, implements algorithms for building and maintaining generative models of the world—internal simulations that allow us to predict future states, imagine counterfactuals, and navigate complex environments. In PTSD and depression, these circuits may implement overly negative or threat-biased generative models, leading to persistent negative expectations and avoidance behaviors. Understanding these symptoms as dysfunctional generative modeling suggests interventions that specifically target model updating—whether through cognitive therapy that challenges negative predictions or through pharmacological approaches that enhance neuroplasticity in these circuits.
Cerebellar-Thalamo-Cortical Effort Computation Emerging evidence from my lab and others points to cerebellar contributions to effort-based decision making and the generation of goal-directed behavior. The cerebellum implements algorithms for motor learning and control, but also appears to contribute to cognitive effort computation—determining how much mental effort to invest in different tasks. In negative symptom states (whether in schizophrenia, depression, or other conditions), dysfunction in cerebellar-thalamo-cortical circuits may implement aberrant effort algorithms, leading to reduced motivation and goal-directed behavior.
Prefrontal-Thalamic Context Maintenance The prefrontal cortex, in dynamic interaction with thalamic circuits, implements algorithms for context maintenance and cognitive control—maintaining representations of current goals, rules, and context while flexibly updating them when circumstances change. Dysfunction in these circuits, whether through developmental disruption or acquired damage, leads to the cognitive control deficits observed across psychiatric conditions.
The Clinical Promise: From Algorithm to Intervention
The power of the Algorithmic Circuit Psychiatry framework lies not in its explanatory elegance, but in its therapeutic implications. By understanding what algorithm is failing and how it's implemented in neural circuits, we can design interventions that target specific computational dysfunctions.
Precision Pharmacology Instead of broad-spectrum drugs that affect entire neurotransmitter systems, we can develop medications that target specific circuit-algorithm dysfunctions. The success of Cobenfy in schizophrenia exemplifies this approach—rather than broadly blocking dopamine receptors, it specifically modulates cholinergic input to thalamocortical circuits, potentially restoring adaptive belief updating algorithms.
Targeted Neuromodulation Deep brain stimulation, transcranial magnetic stimulation, and other neuromodulation approaches can be designed to specifically target the circuits implementing dysfunctional algorithms. Instead of stimulating based purely on anatomy, we can stimulate based on functional understanding of circuit-algorithm relationships.
Algorithmic Psychotherapy Even psychological interventions can be informed by this framework. Cognitive behavioral therapy, for instance, can be understood as algorithm retraining—systematically exposing dysfunctional belief updating or value learning algorithms to corrective experiences that promote more adaptive computational patterns.
Biomarker Development Understanding the algorithmic basis of psychiatric symptoms suggests new approaches to biomarker development. Instead of looking for general markers of "depression" or "schizophrenia," we can develop measures that assess specific algorithmic functions—belief updating flexibility, reinforcement learning parameters, working memory gating efficiency—and use these to guide treatment selection and monitor treatment response.
Research Frontiers: Where the Framework Leads
The Algorithmic Circuit Psychiatry framework opens new research directions that bridge computational modeling, circuit neuroscience, and clinical investigation:
Developmental Algorithmic Psychiatry How do the algorithms implemented by neural circuits develop over time? My lab's work on thalamocortical development suggests that many psychiatric conditions may reflect developmental disruptions in algorithm implementation—not simply altered adult brain function, but aberrant development of the very circuits that implement cognitive algorithms. This has profound implications for early intervention and prevention.
Cross-Species Translation One of the great challenges in psychiatric research is translating findings from animal models to human patients. The Algorithmic Circuit Psychiatry framework provides a solution: algorithms are conserved across species even when specific circuit implementations vary. We can study reinforcement learning algorithms in mice, belief updating in non-human primates, and working memory in humans, confident that we're investigating related computational processes implemented in homologous circuits.
Personalized Psychiatry Individual differences in symptom presentation may reflect individual differences in how algorithms are implemented in circuits. Some patients may have dysfunction primarily in belief updating, others in reinforcement learning, still others in cognitive control. By developing measures that assess algorithmic function at the individual level, we can move toward truly personalized psychiatric treatment.
Why This Framework Matters Now
We stand at an unprecedented moment in neuroscience and psychiatry. Our tools for measuring and manipulating brain circuits have never been more precise. Our computational models have never been more sophisticated. Yet psychiatric treatment remains largely unchanged from decades past, and treatment outcomes remain disappointingly limited.
The missing link is a framework that connects circuit-level understanding to algorithmic insight, and both to clinical intervention. Algorithmic Circuit Psychiatry provides that link.
This is not meant to replace computational psychiatry or circuit psychiatry—both have advanced our field enormously and continue to generate crucial insights. Rather, it's about making explicit the connections between these levels and ensuring that our growing understanding of brain circuits translates into better treatments for psychiatric illness.
In the coming posts, I'll explore this framework in much greater depth:
Deep dives into specific algorithms and their circuit implementations across different psychiatric conditions
Analysis of emerging therapeutics through the lens of algorithmic circuit dysfunction
Exploration of developmental questions: how do algorithmic implementations go awry during development?
Translation challenges: moving from algorithmic insights to clinical applications
Future directions: where this framework leads for the next generation of psychiatric research and treatment
I'll draw extensively from my own research on thalamocortical circuits, sleep and memory, cognitive control, and psychiatric dysfunction. But I'll also synthesize insights from across neuroscience, psychiatry, and computational modeling, working to build a comprehensive framework that can guide our field toward more effective, mechanistically-informed treatments.
If you're interested in how the mind's algorithms are implemented in brain circuits—and how understanding these implementations can transform psychiatric treatment—I invite you to join this exploration.
The future of psychiatry lies not in choosing between computational models and circuit-level interventions, but in understanding how they connect. Algorithmic Circuit Psychiatry is that connection.
About This Substack
I'm Michael Halassa, Professor of Neuroscience and Psychiatry at Tufts University, where I lead a research program investigating the neural circuits underlying cognition and their dysfunction in psychiatric illness. My lab has pioneered approaches for understanding thalamocortical circuits, their role in cognitive control, and their involvement in psychiatric conditions like schizophrenia.
Over the past decade, my work has increasingly focused on bridging circuit-level neuroscience with computational understanding of mental function. This Substack represents my effort to synthesize these perspectives into a coherent framework for understanding and treating psychiatric illness.
I publish research in journals like Nature, Cell, Neuron and Nature Neuroscience, but I believe the most important conversations about the future of psychiatry happen not just in academic venues, but in forums like this where ideas can be explored, debated, and refined.
Welcome to the conversation.