From Molecules to Mind: The Need for a Middle Layer
A fundamental challenge in psychotropic biology is bridging the vast gap between molecular pharmacology (drug X binds to receptor Y) and the resulting changes in subjective experience and behavior. The brain operates as a complex, nonlinear dynamical system, where localized receptor actions can ripple out to alter global patterns of activity and connectivity. To understand these emergent phenomena, the Institute has invested heavily in computational neuroscience. We build detailed in silico models of neural circuits and whole-brain networks to simulate and predict the effects of psychotropic compounds before they are ever tested in a living organism. This approach allows for rapid, low-cost hypothesis testing and can reveal counterintuitive effects that would be difficult to detect experimentally.
Types of Models in Our Computational Toolkit
We employ a hierarchy of models, each with different levels of abstraction and purpose:
- Biophysically Detailed Neuron Models: Using the Hodgkin-Huxley formalism, we simulate individual neurons or small microcircuits (e.g., a cortical column), incorporating various ion channels, receptors, and intracellular signaling pathways. We can then 'apply' a drug by altering the kinetics of specific receptors (e.g., increasing GABA-A channel open time for a benzodiazepine) and observe how this changes the firing patterns of the network. This helps explain side effects like sedation or seizure threshold changes.
- Mean-Field or Neural Mass Models: To scale up to entire brain regions, we use models that treat populations of neurons as a single unit, described by their average firing rate or synaptic activity. These models can simulate the macroscopic signals measured by EEG or local field potentials. They are excellent for studying how drugs that alter excitation/inhibition balance (like ketamine or psychedelics) shift brain rhythms from alpha/beta waves to gamma or theta oscillations.
- Whole-Brain Network Models: At the largest scale, we use structural connectivity data from human diffusion tensor imaging (DTI) to create a 'wiring diagram' of the brain. We then place neural mass models at each node (brain region) and connect them according to the empirical white matter tracts. This allows us to simulate whole-brain dynamics and functional connectivity, as seen in fMRI. We can perturb specific nodes or connections pharmacologically and observe how the perturbation propagates, predicting which networks will become more or less synchronized.
Simulating Psychedelic Action and Antidepressant Effects
Our computational work has been particularly illuminating in two areas:
- Psychedelics and Criticality: We have successfully modeled the action of 5-HT2A agonists like psilocybin. By increasing the gain (responsiveness) of excitatory neurons in our whole-brain model, we simulate the drug's effect of increasing neural excitability. This pushes the simulated brain network from a stable, ordered state (characteristic of normal waking consciousness) towards a 'critical' or even chaotic regime. In this state, network activity becomes highly unpredictable and entropic, with long-range correlations appearing between normally disconnected regions. This in silico result perfectly mirrors the increased functional connectivity and entropy measured in human psychedelic fMRI studies, providing a mechanistic explanation for the phenomenon of 'brain network disintegration' and expanded consciousness.
- Ketamine and Bistability: Modeling the NMDA receptor blockade of ketamine on GABAergic interneurons reveals a disinhibition of pyramidal neurons, leading to a burst of glutamate release. Our network models show this can push cortical circuits into a bistable state, where they alternate between two distinct activity patterns. We hypothesize this bistability underlies the dissociative state and may 'reset' pathological, stable attractor states associated with depressive rumination, allowing new patterns to form.
Predictive Power and Drug Discovery
The ultimate goal is predictive computational psychiatry. We are developing a pipeline where:
- A novel compound's pharmacological profile (binding affinities, receptor kinetics) is entered into the model.
- The model predicts its effects on single neurons, local circuits, and whole-brain dynamics.
- It outputs predictions for changes in specific behavioral domains (mood, cognition, perception) and potential side-effect profiles (e.g., risk of inducing mania or psychosis).
This allows us to virtually screen thousands of hypothetical molecules, prioritizing only the most promising for costly and time-consuming synthesis and animal testing. Furthermore, these models can be personalized by using an individual patient's own brain scan data to create a 'digital twin' of their brain, on which we can test different drug options to predict their unique response. While still a nascent field, computational modeling at the Institute is rapidly becoming an indispensable tool for translating molecular knowledge into a true science of mind-altering drug effects.