The Varieties of the Psychedelic Experience: A Preliminary Study of the Association Between the Reported Subjective Effects and the Binding Affinity Profiles of Substituted Phenethylamines and Tryptamines

Classic psychedelics produce a wide range of subjective effects that depend strongly on set and setting but are also thought to share a common mechanism via partial agonism at serotonin (5-HT) 2A receptors. Despite this commonality, users and some experimental work report phenomenological differences across compounds, including among substituted phenethylamines and tryptamines. Previous explanations for that diversity emphasise set/setting, functional selectivity at 5-HT receptors, or differing affinities for other 5-HT subtypes; Zamberlan and colleagues investigate a complementary possibility that polypharmacology beyond the 5-HT system—including affinities at other neurotransmitter receptors, transporters and ion channels—contributes to the variety of reported subjective effects. This study aims to test whether similarity in binding affinity profiles across a range of targets is associated with similarity in the language used to describe subjective experiences. Using binding affinity data from two independent sources and natural language processing applied to a large repository of self‑reported psychedelic experience narratives, the investigators ask whether compounds with similar affinity “fingerprints” have semantically similar reports, and whether particular receptors or receptor combinations better predict the reported phenomenology. The work is presented as a methodological framework that could be applied to other psychoactive compounds as binding and report data accumulate.

The investigators combined three data streams: (1) published binding affinity measurements (equilibrium dissociation constants K i, transformed to pK i = -log10(K i)) for a set of psychedelic compounds assayed across up to 42 binding sites; (2) structural similarity computed from 2D atom-pair fingerprints with the Tanimoto coefficient; and (3) semantic representations of subjective reports drawn from the Erowid Experience Vaults. Two independent binding datasets were used: a primary PDSP-derived set covering 18 compounds and 42 sites, and a replication set covering 19 compounds and 14 sites. Compounds were selected if they had binding affinity data at a sufficient number of sites and at least 10 subjective reports in the Erowid corpus; this resulted in 18 compounds for the main analysis and 19 for replication. Binding affinity matrices were normalised with the procedure of earlier PDSP work, log-transformed and sign-inverted to yield pK i values; pairwise similarity of compounds' affinity profiles was computed as Pearson correlation across sites where both compounds had measurements. Structural similarity used ChemMine atom-pair Tanimoto scores on SMILES-derived representations. For the subjective reports, the team tokenised, lemmatised and filtered the Erowid narratives, removed substance names and route terms, and retained words present in more than 5% and less than 95% of documents, yielding a vocabulary of 1,466 words. Latent Semantic Analysis (LSA) was applied to the tf-idf weighted word-by-document matrix and reduced via singular value decomposition, retaining D = 20 dimensions; semantic similarity between compounds was computed as Pearson correlation between columns of the rank-reduced matrix. Principal component analysis (PCA) was subsequently applied to the LSA-reduced matrix to extract five principal components of the narratives (the first five components explaining ≈58% of variance) and to obtain per-compound component scores. The authors also implemented an in silico ‘‘virtual receptorome’’ screening to estimate binding affinities from public databases (ChEMBL, KiDatabase) using structure-based inverse distance weighting of known ligands (fingerprint similarity via RDKit, Tanimoto distance) implemented in KNIME; an optimisation procedure (power parameter p = 4.0, ‘‘Layered’’ descriptor) was benchmarked on known psychiatric drugs. To evaluate which subsets of receptors best predict semantic similarity, an exhaustive search over receptor combinations was performed with the constraint that each combination include at least one of 5-HT, dopamine, adrenergic or muscarinic groups, and robustness was assessed via bootstrap (1,000 iterations) in which LSA was re-run on random half-samples of words.

Semantic, pharmacological and structural similarities showed positive associations. Using the primary PDSP-derived data (18 compounds, 42 sites), the correlation between binding affinity profile similarity and reported subjective-effect semantic similarity was R = 0.60 (p = 2.2e-16). Binding affinity profile similarity also correlated with molecular structure similarity (R = 0.46, p = 2.5e-9), and semantic similarity correlated with molecular structure similarity (R = 0.30, p = 1.9e-4). The virtual screening (in silico estimates of binding affinities) correlated significantly with empirical PDSP affinities (R = 0.42, p = 7.3e-8), supporting the plausibility of the empirical affinity data. Replication with an independent dataset (19 compounds assayed at 14 sites) produced qualitatively similar results: semantic and binding-affinity similarity matrices resembled those from the main dataset, with reported correlations typically R > 0.4 across a range of LSA dimensions; an example value reported was R = 0.48 (p = 1e-11) for one comparison. Optimisation over receptor subsets showed that predictive power increased when considering targets beyond 5-HT receptors. The combination of sites that maximised prediction of semantic similarity comprised 5-HT, dopamine (D), opioid, muscarinic (M), NMDA receptors and the Ca2+ channel, yielding R = 0.63 (p < 1e-17). The best prediction of molecular-structure similarity from binding profiles used 5-HT, imidazoline, σ and muscarinic receptors (R = 0.52, p < 1e-11). Bootstrap analysis (1,000 iterations, LSA on 50% random word samples) showed these results were robust: certain receptor types were selected frequently across iterations, notably 5-HT, dopamine, opioid, muscarinic, cannabinoid and NMDA receptors together with the Ca2+ channel. PCA of the LSA-reduced word-by-document matrix yielded five principal components explaining ≈58% of variance. The components were labelled by the authors as: 1) ‘‘perception’’ (≈23% variance; top words: visuals, color, visual, pattern, saw, reality, face, their, outside, vision), 2) ‘‘body load’’ (≈13%), 3) ‘‘preparation’’ (≈9%), 4) ‘‘dependence’’ (≈7%), and 5) ‘‘therapeutic’’ (≈5%). Per-compound scores showed expected patterns (for example, high ‘‘perception’’ scores for many classic psychedelics except ibogaine; higher ‘‘dependence’’ and ‘‘therapeutic’’ scores for ibogaine and MDMA). Correlating receptor affinities with component scores identified plausible links: higher affinity at some 5-HT receptor subtypes (including 5-HT2A) correlated positively with the ‘‘perception’’ component, while affinities at transporters (SERT, DAT), opioid receptors (MOR, KOR) and NMDA receptors showed generally negative correlations with the ‘‘therapeutic’’ component; Ca2+ channel affinity correlated positively with the ‘‘dependence’’ component. The authors present radar plots illustrating these receptor–component relationships, including examples for 5-HT2A, MOR and Ca2+.

Zamberlan and colleagues interpret their findings as preliminary evidence that similarity in binding affinity profiles across a broad set of receptors, transporters and ion channels is associated with semantic similarity in self‑reported psychedelic experiences. They emphasise that this result does not negate the central role of 5-HT2A agonism — which remains a necessary condition for classic psychedelic effects — but suggests that polypharmacology beyond 5-HT2A and functional selectivity likely act together to shape qualitative differences between compounds. The authors position their results relative to prior literature on functional selectivity and cross-tolerance: while differential intracellular signalling downstream of 5-HT2A may account for some diversity, the observed correlations imply that affinities at other systems (notably dopamine receptors, muscarinic and opioid receptors, NMDA receptors and the Ca2+ channel) also relate to distinct facets of reported experience. They note specific examples consistent with earlier work, such as dopaminergic actions of LSD and potential glutamatergic involvement linking some psychedelics to NMDA-antagonist phenomenology. Several limitations acknowledged by the authors temper their conclusions. The subjective data derive from self‑selected, anonymous Erowid narratives that lack laboratory verification of substance identity, dose and context; they may contain expectation effects and variable user histories. Binding-affinity data have important constraints: K i values do not indicate functional efficacy (agonist versus antagonist), radioligand choice can bias estimates towards particular receptor conformations, data are incomplete for some compounds (missing values), and in vitro affinity does not map directly onto in vivo potency because of pharmacokinetics, metabolism and tissue distribution. The authors also caution that their Fourier-analogue decomposition of subjective effects is limited by the brain's non-linearity and by interactions among neuromodulatory systems. Given these caveats, the investigators treat their results as preliminary but useful: the methodological pipeline—combining natural language processing of large narrative corpora with binding-affinity and structural data—provided sensible, reproducible patterns across two datasets. The paper calls for follow-up work including replication of binding assays, functional-effector assays (PLC, PLA2, β-arrestin, etc.), standardised psychometric and neuroimaging studies, more complete affinity screens, and the use of in silico modelling to complement empirical data. Finally, the authors suggest the approach could be extended to study therapeutic and adverse effects of neuropsychiatric medications by linking semantic report similarity to pharmacological and brain-wide measures, enabling a kind of reverse engineering of desirable and undesirable affinity profiles.