K.H.A.I.T.O.

Brain condition

Present a stimulus

View

Layout

Color by

Display options

Pull apart0% Impulse density1.0×

Region activity bar = now · tick = healthy baseline

normal activity (tissue colour) above baseline below baseline excitatory impulse inhibitory impulse compound in tissue blocked at blood–brain barrier

Drag to rotate · scroll or pinch to zoom · point at a region to see its name

K.H.A.I.T.O.

Kinetic Hybrid AI for In-silico Toxicity Optimization

Compound lab

MW (Da)

cLogP

TPSA (Å²)

H-donors

Charge

Blood–brain barrier crossing–

Pharmacokinetics label PET lit est gene

Dose

Dose1 mg Body weight70 kg

Clearance scales with body weight^0.75 (allometric scaling, Anderson & Holford 2008). Not valid under 2 years, when liver enzymes are still maturing.

Exposure

Spread timer one dose · from published t½, Tmax, F, Kp,uu

Administer a dose to time how fast it reaches and leaves the brain.

Target engagement

Administer a dose to see target engagement.

Where it acts Δ activity vs. untreated

No compound in the brain yet.

Evidence for this prediction

Accuracy check model vs. published PET

Data sources

PTSD and anxiety models: Etkin and Wager 2007, Am J Psychiatry, a voxelwise meta-analysis of emotional-processing imaging in PTSD, social anxiety disorder and specific phobia; and Wang et al. 2016, Sci Rep, a resting-state meta-analysis of 15 studies and 286 people with PTSD.
Huntington's model: striatal D1, D2/D3 and dopamine-transporter loss and cortical hypometabolism, reviewed in Cybulska et al. 2020, Molecules 25:482 and Roussakis and Piccini, J Huntingtons Dis 2015; microglial activation from Pavese et al. 2006, Tai et al. 2007, Politis et al. 2015 and Roussakis et al. 2023, Brain Commun (15 patients, 21 controls).
Migraine model: dorsal pons activation during attacks (Weiller et al. 1995, Nat Med; Borsook and Burstein 2012, Cephalalgia) and hypothalamic activation across the migraine cycle (May and Burstein 2019, Cephalalgia; Schulte and May 2016, Brain).
Insomnia model: the hyperarousal account of Nofzinger et al. 2004, Am J Psychiatry, from waking-to-sleep glucose metabolism.
Plant and natural compounds: trial results are taken from the published record and are shown in the Evidence panel with the result as reported, including the negative ones — Snitz et al. 2009 and DeKosky et al. 2008, JAMA (Ginkgo, 3,069 participants, negative); Brimson et al. 2021, Sci Rep (Bacopa meta-analysis, 645 completers, mostly null); Hausenblas et al. 2013, J Integr Med (saffron); Linde et al., Cochrane (St John's wort); Lopresti et al. 2014 and Sanmukhani et al. 2014 (curcumin); Sarris 2018, Phytother Res (ashwagandha and others); de la Torre et al. 2016, Lancet Neurol (EGCG). Extract properties are those of a single named constituent, because an extract is a mixture and this model takes one molecule at a time.
Brain anatomy: AAL atlas (Tzourio-Mazoyer et al. 2002, NeuroImage) and MNI152 template (ICBM 2009), from the NiiVue demo images on GitHub.
Receptor and transporter maps are all open-licence. Three are measured images resampled here at every brain point: D2/D3 from [11C]raclopride PET in 156 adults (Malén et al. 2022, CC0), GABA-A α5 from [11C]Ro15-4513 PET in 24 adults (Lukow et al. 2022, CC BY 4.0) and microglial TSPO from [11C]PBR28 PET (Lois et al. 2018, MIT). Six more — NMDA, mGluR5, SERT, 5-HT2A, mu-opioid and M1 — have no open image, so their 16 regional values are typed from open-access CC BY papers and spread smoothly across the brain points for display: the gradient you see inside a region is interpolation, not measurement. Where a paper did not report a region, the value is the model’s own estimate and is named on the map. The dopamine transporter borrows the shape of the CC0 D2/D3 image, since both mark dopamine terminals. Cortical gaps can also be filled from Zilles and Palomero-Gallagher 2017 autoradiography (CC BY 4.0). Tracer, sample size and licence are shown on each compound.
The six targets with no PET image of their own no longer use the Allen atlas. Synaptic density is now [11C]UCB-J PET in 18 adults (Cayir et al. 2026, Brain Commun, CC BY 4.0), a measured scan. GABA-B uses [3H]CGP54626 autoradiography of eight cortical areas (Zilles and Palomero-Gallagher 2017, CC BY 4.0) with the deep structures scaled from RNA. Adenosine A1, the oxytocin receptor and NKCC1 use brain-region RNA from the Human Protein Atlas (proteinatlas.org; Uhlén et al. 2015, Science 347:1260419, CC BY 4.0), which reports one value for the whole cortex, so the nine cortical regions are equal for those three and a real cortical gradient cannot be shown. Acetylcholinesterase is the model’s own estimate ordered to match [11C]MP4A and [11C]PMP PET of enzyme activity, because the only open data is RNA and RNA marks where the gene is transcribed rather than where the enzyme works. RNA is a proxy for protein, not the protein itself.
3D brain surface: Conte69 32k mid-thickness cortical surface (Van Essen et al. 2012, Cereb Cortex) with Desikan–Killiany parcels, and subcortical nucleus surfaces, from the ENIGMA Toolbox (Larivière et al. 2021, Nat Methods). Cerebellum, brainstem and hypothalamus are smoothed from the atlas points. Head: scalp surface of the MNI ICBM152 T1 template (Fonov et al. 2011, NeuroImage). Arteries follow the main cerebral artery routes approximately. The Conte69 surface comes from Human Connectome Project open-access data and is used under the WU-Minn HCP Open Access Data Use Terms, which require that it be passed on under those same terms and acknowledged: “Data were provided in part by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University.”
Cortical pathway strengths: group-average diffusion-MRI tractography in 50 healthy adults, MICA-MICs (Royer J, et al. 2022, Sci Data 9:569; OSF project j532r, CC0 1.0), Desikan–Killiany parcels pooled into the eight cortical regions and corrected for region size. The HCP connectome previously used here has been removed.
Autism connectivity: ABIDE I has been removed. The pathway change now follows one published, open-licence result — 23 autistic and 20 non-autistic adults, right-thalamus seed (Transl Psychiatry 11:93, 2021, Table 3, CC BY 4.0), which found the thalamus over-connected with motor, somatosensory and superior parietal cortex (cluster p = 0.007–0.012) and no difference for default-mode or executive-control seeds. The paper gives cluster p values and not effect sizes, so the direction is measured and the size is the model’s own estimate. The rest of the autism model (local excitation, amygdala and sensory gain, fusiform response, Purkinje loss, NKCC1 sensitivity, microglial pattern) is unchanged and comes from the papers cited above.
Disease maps (cortical-thickness Cohen's d, case minus control) read from the published ENIGMA working-group papers and cited as findings (the ENIGMA Toolbox, Larivière et al. 2021, Nat Methods, states no licence on its data files): autism, ADHD, Parkinson's, temporal-lobe epilepsy and schizophrenia working-group summary statistics. Cerebral palsy follows the SCPE MRI Classification distribution (Himmelmann et al. 2021, Front Neurol); Alzheimer's follows the published atrophy pattern. Inflammation patterns follow TSPO PET studies cited per condition.
MSC-exosome behaviour: Perets et al. 2019 (Nano Lett), Long et al. 2017 (PNAS), Drommelschmidt et al. 2017 (Brain Behav Immun), Chen et al. 2020 (Cell Death Dis) and the Xie et al. 2023 phase I/II Alzheimer's trial (Gen Psychiatr). Biodistribution from Wiklander et al. 2015 (J Extracell Vesicles).
Depression and OCD models: ENIGMA cortical-thickness summary statistics via the ENIGMA Toolbox — depression 1,902 patients vs 7,657 controls (Schmaal et al. 2017, Mol Psychiatry), OCD 1,497 vs 1,435 adults (Boedhoe et al. 2018, Am J Psychiatry) — with the functional pattern from Mayberg et al. 1999/2005, Koenigs and Grafman 2009, Siegle et al. 2007, Pizzagalli et al. 2009, Saxena and Rauch 2000, Menzies et al. 2008. Inflammation from TSPO PET: Setiawan et al. 2015 (depression), Attwells et al. 2017 (OCD).
Tourette model: Kalanithi et al. 2005, PNAS and Kataoka et al. 2010 (loss of striatal inhibitory interneurons); Lerner et al. 2012, Brain (reduced GABA-A binding on PET); Williams et al. 2021, Mol Psychiatry (ENIGMA-Tourette cortical thickness); Kumar et al. 2015 (microglial activation). No public ENIGMA data file exists for Tourette, so these values were typed from the papers.
Stroke model: Baron et al. 1981 (crossed cerebellar diaschisis); Grefkes and Fink 2014 (network recovery); Gerhard et al. 2005 and Thiel et al. 2010 (TSPO PET after stroke); Xin et al. 2013 and Doeppner et al. 2015 (MSC exosomes in rodent stroke).
Addiction model: lower striatal D2/D3 availability and blunted dopamine release (Volkow et al. 1993, 2014); lower prefrontal and cingulate metabolism (Volkow et al. 1992, 2011); cue reactivity meta-analysis (Kuhn and Gallinat 2011); microglial activation in methamphetamine use disorder (Sekine et al. 2008); naltrexone mu-opioid blockade (Weerts et al. 2008) and its effect on cue-induced craving (Monterosso et al. 2001).
Blood–brain barrier data and AI predictor training set: B3DB (Meng et al. 2021, Sci Data; github.com/theochem/B3DB). Molecular properties computed with RDKit.

References and values were compiled from published sources by the model author. Check them before citing.

Licences and credits

Every map in this model, what it may be used for, and the credit its licence requires. Open-licence maps were resampled here from the published images.

Map	Source	Licence	Commercial use
Dopamine D2/D3	Malén T, Santavirta S, De Maeyer S, et al. 2022, NeuroImage 255:119149. [11C]raclopride PET, 156 adults. Via neuromaps.	CC0 1.0 (public domain)	Yes, no conditions
GABA-A α5 subunit	Lukow PB, Martins D, Veronese M, et al. 2022, J Neurosci. [11C]Ro15-4513 PET, 24 adults.	CC BY 4.0	Yes, with the credit shown here
Microglia (TSPO)	Lois C, González I, Izquierdo-García D, et al. 2018, J Nucl Med. [11C]PBR28 PET. github.com/hookerlab	MIT	Yes, with credit
Brain parcels	AAL atlas, Tzourio-Mazoyer N, et al. 2002, NeuroImage 15:273; Rolls ET, Huang C-C, Lin C-P, et al. 2020, NeuroImage 206:116189 (AAL3).	Free for research and commercial use	Yes
Template space	MNI ICBM152 2009 non-linear asymmetric, Fonov V, et al. 2011, NeuroImage 54:313. Copyright 1993–2009 Louis Collins, McConnell Brain Imaging Centre, Montreal Neurological Institute.	Free with the copyright notice kept	Yes
Blood–brain barrier data	B3DB, Meng F, Xi Y, Huang J, Ayers PW 2021, Sci Data 8:289. github.com/theochem/B3DB	CC BY 4.0	Yes, with credit
Resampling toolbox	neuromaps, Markello RD, Hansen JY, Liu Z-Q, et al. 2022, Nat Methods 19:1472.	BSD-3	Yes, with credit
NMDA, mGluR5, SERT, 5-HT2A, mu-opioid, M1	Regional values typed from open-access CC BY papers: Beck et al. 2021 Mol Psychiatry (NMDA); Milella et al. 2025 (mGluR5); Kim et al. 2024 Int J Mol Sci 26:252 (SERT, 5-HT2A); Noppari et al. 2025 EJNMMI (mu-opioid); Volpi et al. 2026 (M1). Cortical gaps where a paper reports none are the model’s own estimate and are named on each map.	CC BY 4.0	Yes, with the credit shown here
Dopamine transporter	No open image exists. The shape is borrowed from the CC0 D2/D3 image (Malén et al. 2022) because both mark dopamine terminals; the striatal anchor is Crine et al. 2026 (CC BY 4.0).	CC0 + CC BY	Yes
Cortical autoradiography (gap filling)	Zilles K, Palomero-Gallagher N 2017, Front Neuroanat. 44 cortical areas, fmol/mg protein.	CC BY 4.0	Yes, with credit
Synaptic density (SV2A)	Cayir S, et al. 2026, Brain Commun. [11C]UCB-J PET, 18 adults; cortex from Luoma et al. 2025.	CC BY 4.0	Yes, with credit
GABA-B, adenosine A1, oxytocin receptor, NKCC1	Zilles and Palomero-Gallagher 2017 autoradiography for GABA-B cortex; brain-region RNA from the Human Protein Atlas (proteinatlas.org; Uhlén MJ, et al. 2015, Science 347:1260419) for the rest.	CC BY 4.0	Yes, with credit
Acetylcholinesterase	The model’s own estimate, ordered to match [11C]MP4A and [11C]PMP PET (Kuhl et al. 1999, Neurology; Herholz et al. 2004). Not copied from any dataset.	None	Yes
Cortical pathway strengths	MICA-MICs group-average structural connectome, 50 adults. Royer J, et al. 2022, Sci Data 9:569; OSF project j532r.	CC0 1.0 Universal	Yes — citation requested, not required
Autism connectivity change	Transl Psychiatry 11:93 (2021), Table 3 — direction only, in an independent cohort of 23 autistic and 20 non-autistic adults. Sizes are the model’s own estimate. No ABIDE data is used.	CC BY 4.0	Yes, with credit
Disease cortical-thickness maps	Published group summary statistics (Cohen’s d, case minus control) from the ENIGMA working-group papers: van Rooij et al. 2018 (autism), Hoogman et al. 2017/2019 (ADHD), Laansma et al. 2021 (Parkinson’s), Whelan et al. 2018 (epilepsy), van Erp et al. 2018 (schizophrenia), Schmaal et al. 2017 (depression), Boedhoe et al. 2018 (OCD), Williams et al. 2021 (Tourette). The ENIGMA Toolbox licenses its code under BSD 3-Clause but states no licence on its data files, so these numbers are read from the papers and cited as published findings rather than redistributed as a dataset.	Published numeric findings, cited	Yes, with citation
Cortical surface (the 3D brain you see)	Conte69 32k mid-thickness surface, Van Essen DC, et al. 2012, Cereb Cortex 22:2241, via the ENIGMA Toolbox.	WU-Minn HCP Open Access Data Use Terms	Only if those terms are passed on

One row is still marked no, and it is not about non-commercial data. The WU-Minn HCP Open Access Data Use Terms let anyone redistribute HCP data “and any derived data… as long as the data are redistributed under these same Data Use Terms”, and say nothing either way about commercial use. The 3D surface may therefore be passed on, but only together with those terms, which whoever receives this application has to accept in turn. Everything else is open-licence, the model’s own estimate, or a published number cited rather than redistributed. One residual question: the Human Protein Atlas is CC BY 4.0 and states that commercial use is encouraged, but part of its tissue RNA may come from GTEx, whose own terms could not be read.

Prototype v0.9.8 — a teaching model. Brain anatomy comes from the AAL atlas, receptor and transporter density from human PET tracers, now all on open licences — three as measured images sampled at every brain point, the rest as regional values typed from open-access papers and smoothed for display, as the Licences and credits table sets out, extra molecular targets from open-licence PET, autoradiography and Human Protein Atlas RNA, cortical wiring from the CC0 MICA-MICs tractography connectome, the disease models from ENIGMA group summary statistics and the published literature, blood–brain barrier values from B3DB, exosome behaviour from preclinical and early-phase studies, and drug parameters from labels and PET studies — each value tagged by source. The activity dynamics, the way receptor occupancy and exosome cargo are turned into activity change, the disease and inflammation patterns, the trophic (repair) effect, and every value tagged "est" are simplified and illustrative. Gene expression is a proxy for protein density, not the protein itself. The GABA-A map now measures the α5 subunit, which sits more in the limbic regions than the general benzodiazepine site a benzodiazepine acts on, so benzodiazepine predictions are weighted too far towards the hippocampus and amygdala. The Evidence panel names the published work behind each step of a prediction and says when no study tests that compound in that condition. The addiction model, the craving index and the reward cue are an illustration of published group findings, not a measure of craving in any person. The accuracy check covers only drug target occupancy, not effects on behaviour, craving or any exosome effect. The epilepsy model bursts rather than settling, so what is reported for it is the typical height of a burst averaged over many bursts, not the activity at one moment. This is an educational simulation: it is not validated for any decision and does not predict real drug or exosome effects in a person. Do not use it for clinical or experimental decisions.

Most improved

K.H.A.I.T.O.