The recreational dose which 2C-B is typically taken at ranges between 10 to 30 mg. The effects in this dose range include only non-existent to mild alterations in perception or cognition, which would be the signature of 5-HT2a agonism. Instead the reported effects of 2C-B mainly include stimulation and induced euphoria. The 5-HT2c receptor is believed to be involved in mediating the release of norepinephrine and dopamine, which are associated with stimulation and positive affect, among other things. Given its high affinity for the 5-HT2c receptor, we believe that this is likely responsible for driving the commonly reported subjective effects of 2C-B. Notably, this pattern of neuromodulator release is similar to that of MDMA, which 2C-B is sometimes compared to at low doses. At dose ranges above 30 mg, the effects of 2C-B are reported to become increasingly intense and uncomfortable, as well as to take on a more traditional psychedelic profile. We believe this corresponds to increased 5-HT2a agonism with increase in dose.

5-MeO-DMT is a highly biased 5-HT1a agonist. According to our model, we expect biased 5-HT1a agonists to be anxiolytic, yet the 5-MeO-DMT experience seems to be quite intense. Our model predicts that this intensity comes from the extremely high dose at which it is taken. This not only results in a high level of 5-HT2a agonism in addition to the 5-HT1a agonism, but also off-target serotonin effects which disrupt thermoregulation and other bodily functions. Subjective reports indicate that at low doses 5-MeO-DMT acts as an anxiolytic in a manner similar to other 5-HT1a agonists such as buspirone or 8-OH-DPAT. In our model, we predict that only at very high doses will biased 5-HT1a agonists provide long-term therapeutic effects, and this is what seems to be the case for 5-MeO-DMT. As such, we recommend exploring the space of more balanced 5-HT agonists such as 5-MeO-MIPT or 5-MeO-DIPT, which may have long-term efficacy at lower doses and thus are likely to be more psychologically tolerable.

In this project we were interested in studying the effects of serotonergic psychedelics on cortical dynamics, and the dynamics of the prefrontal cortex (PFC) in particular. In the PFC there is a large population of pyramidal neurons which express both 5-HT1a and 5-HT2a receptors. In this cell population 5-HT1a agonism exerts an inhibitory effect, and 5-HT2a agonism exerts an excitatory effect. The population of these pyramidal cells collectively computes the high-level representations encoded by the PFC. These cells' connectivity structure induces the energy landscape which we are modeling. Given the excitatory effects of 5-HT2a, we model it as injecting structured noise into the energy landscape. Given the inhibitory effects of 5-HT1a, we model it as performing a global smoothing of the energy landscape. These two choices closely relate to the use of Hamiltonian Monte Carlo sampling when performing inference on energy functions reflecting neural data. In this work we make the explicit connection between this sampling algorithm and the neuromodulatory effects of 5-HT.

We start with the assumption that cortical networks are performing some form of hierarchical predictive processing. This means that a given population of neurons is trying to predict (and thus model) the distribution of incoming signals from upstream levels of the hierarchy. In our energy based model framework, this corresponds to trying to learn an energy function which is able to match a target energy function that captures the true distribution of possible upstream signals to be predicted. We further assume that these prediction errors are both felt as unpleasant and indicate behavioral maladaptivity.

Given these assumptions, the goal of learning should be to minimize the distance between the distribution of neural activity induced by the current energy landscape and the target distribution. We measure this distance using KL-divergence. The expectation is that 5-HT neuromodulation should aid the learning process, and thus the effectiveness of a drug can be measured by how much it contributes to reducing KL-divergence as compared to a baseline. We use this as our measure of therapeutic efficacy. The introduction of neuromodulation also disrupts the current energy function, and can result in transient increases in KL-divergence during the acute drug phase. In our model this would correspond to an increase in psychological distress. We therefore use the cumulative stepwise increase in KL-divergence during the experiment as a measure of the overall psychological tolerability of the drug intervention, with lower values corresponding to greater tolerability.