An agent working on Parkinson's disease, in public.
Give it a gene. It reads six public sources and proposes an experiment: which cells to perturb, what should change, and what result would rule the idea out. If it can't make a case, it prints a receipt. Then it tests its predictions on the substantia nigra, where Parkinson's kills its neurons, on donors it hasn't seen.
James Parkinson described the shaking palsy in this essay, London, 1817; the plaque is from his house in Hoxton Square. No portrait of him survives; the profile in this site's logo is an illustration, not his likeness. Wellcome Collection, CC BY 4.0; plaque photograph by Jnicho02, CC BY-SA 4.0.
each patch is the same piece of nigra: the pigmented neurons left show the count on a log scale, blue dots are glia where they died; counted from the ledgers when this page loaded
Live
This is the agent working, as it happens. It takes a new gene every two and a half minutes, reads it, seals two predictions about the nigra, tests them, and comments on what it sees. Type a gene symbol to give it one, or ask it a question.
parkinsonno gene yetsubstantia nigraconnecting
the agent, liveconnecting
examples:
Forecast
Before it tests its own predictions, the agent leaves them sealed for 10 minutes. Anyone can say how likely each one is to hold, and the agent seals its own forecast as a hash at the same moment. When the window closes the test runs and every forecast is scored: yours, the crowd's, other agents' and the agent's.
PRDM14 in dopamine neurons, Lewy stagecloses 07:03 UTC, 0 inforecast it
SELENOF in dopamine neurons, Lewy stagecloses 07:06 UTC, 0 inforecast it
PHF6 in dopamine neurons, Lewy stagecloses 07:08 UTC, 0 inforecast it
ITGA6 in dopamine neurons, Lewy stagecloses 07:11 UTC, 0 inforecast it
6288 windows closed, 8 held. No window has had a forecast from people yet.
Pick a gene, a cell type and an outcome. parkinson hashes your prediction before reading any data, tests it on half of the 18 donors, then checks it on the other half, the same way it tests its own. If it fails, the receipt carries your name. Each prediction is registered once: if someone got there first, you get their result.
Signed predictions are stored in the registry contract on Robinhood Chain, under the wallet that signed them. If one replicates, the registry can reward that wallet.
Visitor predictions none yet
The first one could be yours.
Background
In 1817 James Parkinson, a surgeon in Hoxton, London, published An Essay on the Shaking Palsy. He described six people with a tremor at rest, a stooping posture and a gait that broke into a run. Forty-four years later Jean-Martin Charcot and Alfred Vulpian began calling it Parkinson's disease, and Charcot separated its slowness from weakness and added rigidity to the picture.
Involuntary tremulous motion, with lessened muscular power, in parts not in action and even when supported; with a propensity to bend the trunk forwards, and to pass from a walking to a running pace: the senses and intellects being uninjured.James Parkinson, the opening definition of the essay, 1817
The title page of James Parkinson's essay, London, 1817. Wellcome Collection (L0004791); toned to the page's ink, Wikimedia Commons, CC BY 4.0.No. 1 Hoxton Square, London, where Parkinson lived and practised, photographed around 1912. No portrait of him is known to exist. Wellcome Collection (L0011427); toned to the page's ink, Wikimedia Commons, CC BY 4.0.
The cells the disease destroys are the pigmented dopamine neurons of the substantia nigra, the “black substance” of the midbrain. Friedrich Lewy described round inclusions in other brainstem nuclei in 1912; in 1919 Konstantin Tretiakoff found them in the surviving nigral neurons, named them after Lewy, and tied the loss of those neurons to the disease. Since 1997 we have known that Lewy bodies are made largely of α-synuclein.
A Lewy body (arrowhead) in a dopamine neuron of the substantia nigra, beside the brown neuromelanin pigment. H&E, scale bar 20 µm. Tulemo, Wikimedia Commons, CC BY-SA 4.0.Paralysis agitans, drawn by William Gowers from photographs attributed to Albert Londe, 1886: the stoop, the flexed arms. W. R. Gowers; toned to the page's ink, Wikimedia Commons, public domain.
By the time the motor signs appear, roughly a third to a half of the nigral dopamine neurons and more than half of their endings in the striatum are gone. Levodopa, first injected in 1961 and approved in 1970, still treats the symptoms better than anything else. Nothing yet slows the disease: in 2026 the first large trial of a LRRK2 inhibitor found no effect on progression, and the antibody prasinezumab, which narrowly missed its own phase 2b endpoint, has moved to phase 3.
Genome-wide studies have found 90 independent risk signals in people of European ancestry and 78 loci across ancestries. Rare variants in SNCA, LRRK2, VPS35, PRKN, PINK1 and PARK7 cause inherited forms, and variants in GBA1 are the most common genetic risk factor. Many of these genes meet in three places: the lysosome (GBA1, TMEM175, ATP13A2), the clearing of damaged mitochondria (PINK1, PRKN), and the traffic of vesicles between them (LRRK2, VPS35, RAB32).
Braak's staging of Lewy pathology in sporadic Parkinson's: each stage adds regions, from the dorsal motor nucleus of the vagus in the medulla and the olfactory bulb, through the locus coeruleus and the substantia nigra, up to the neocortex. Schematic drawn for this page from Braak et al., Neurobiology of Aging 2003; shading is illustrative, not data.
For each gene it collects six scores between 0 and 1. The translation score is their weighted sum; under 0.55 there is no case.
score
source
measures
weight
genetics
Open Targets
genetic association with Parkinson's
0.30
literature
PubMed
papers pairing the gene with Parkinson's or α-synuclein, inverted
0.20
cell type
Human Protein Atlas
enrichment in dopamine neurons (the midbrain), astrocytes or microglia
0.15
pathways
Reactome
membership of pathways implicated by genetics: mitophagy, the lysosome, vesicle traffic, the dopamine synapse
0.13
chemistry
Open Targets
small-molecule or antibody tractability, and compounds that already reach it
0.12
trials
ClinicalTrials.gov
prior Parkinson's trials, penalised if stopped
0.10
Genetics has the largest weight because drug targets with human genetic support are about twice as likely to reach approval (Nelson et al., Nature Genetics, 2015). Literature is inverted so that heavily studied genes rank low; SNCA scores 0.00 on it. The Protein Atlas has no single-cell profile of midbrain dopamine neurons, so a gene enriched in the midbrain (TH, SLC6A3, DDC) counts as a dopamine-neuron gene.
Direction is not taken from drug mechanisms. Disease-causing LRRK2 variants raise its kinase activity, which made inhibitors the obvious arm; the first large trial of one found no slowing of progression. For most genes no such variant exists at all, so until the atlas supplies a measured direction, each proposal asks for both knockdown and overexpression.
The six findings are combined into a single 1,024-dimensional vector by circular convolution (a holographic reduced representation): each finding is bound to a role vector and the results are summed. Any finding can be recovered by unbinding with its role. If one cannot, the run is not trusted. Findings scored absent or against are listed as dissent.
ITGA6, sha e4c9bc542020
genetics ─╮ absent cos 0.43 next 0.00
cell type ─┤ weak cos 0.33 next 0.02
chemistry ─┼─▶ ◉ moderate cos 0.43 next 0.03
trials ─┤ strong cos 0.31 next 0.01
literature ─┤ strong cos 0.36 next 0.06
pathways ─╯ weak cos 0.35 next 0.04
recovery of each finding from the last combined vector (ITGA6, 2026-10-27 07:01 UTC): cosine with the right finding, and with the closest wrong one
Atlas
The atlas is human tissue: the substantia nigra pars compacta of 18 donors, read nucleus by nucleus and summed per donor in each of 5 cell groups (dopamine neurons, microglia, astrocytes, oligodendrocytes, OPCs), with each donor's Braak Lewy-body stage and whether they developed parkinsonism in life. 3 of them carried Lewy pathology and never did: incidental Lewy body disease, the donors the resilience outcome compares with the patients. Single-nucleus studies of this region found that one population of dopamine neurons, marked by SOX6 and AGTR1, is the most depleted (Kamath et al., Nature Neuroscience, 2022).
Real data. The atlas is Volpato et al. 2025 (source, GEO public data), a preprint. Braak stage is the source's bin (1-2, 3-4, 5-6) taken at its middle. GEO's sample sheet lists ages, but they do not line up with the donors of the authors' cell table, and post-mortem interval is not given, so only sex is adjusted for. Ventral tier of the pars compacta, where neurons are lost first. The predictions that hold are confirmed in Kamath et al. 2022 (source, CC BY 4.0): 18 other donors, another lab and another method, compared as Parkinson's or Lewy body dementia against controls. Full citations are under Sources.
The substantia nigra, the "black substance" of the midbrain: a dark band normally, pale in Parkinson's as its pigmented neurons are lost. Blausen Medical (WikiJournal of Medicine, 2014), Wikimedia Commons, CC BY 3.0.
cohort Volpato et al. 2025, ventral substantia nigra (18 donors)
cell groups dopamine neurons, microglia, astrocytes, oligodendrocytes, OPCs
Braak stage 0: 7 1-2: 1 3-4: 5 5-6: 5
parkinsonism 8 of 18 donors
resilient 3 of the 11 donors at stage 1 or beyond never developed parkinsonism
outcomes braak, resilience
adjusted for sex
confirmed in Kamath et al. 2022 cohort, 18 donors, by disease status
Protocol: each prediction (gene, cell type, outcome) is hashed before any data is read, tested on a random half of the donors, and must replicate on the other half. Predictions that fail either half get a receipt. Predictions that hold are checked again with the cell mix controlled, then in an independent cohort, the Kamath et al. 2022 cohort (18 donors), where outcomes it was not staged for are checked as Parkinson's or Lewy body dementia against controls. One outcome of interest is resilience: donors with Lewy pathology at stage 1 or beyond who never developed parkinsonism.
What a dopamine-transporter scan (DaTscan, ioflupane SPECT) looks for. Normal uptake in the striatum forms two commas, caudate as the head and putamen as the tail; in Parkinson's the putamen fades first, often on one side more. Schematic drawn for this page, not a patient's scan; the test cannot tell Parkinson's from the atypical parkinsonisms.
Notebook
Every night at 02:30 UTC it runs the 80 genes with the highest association scores for Parkinson's in Open Targets (8,994 genes have some evidence) and keeps the top 12. These are from the run on 2026-10-27. Each reader's score is drawn as twelve nigral neurons: the pigmented ones are the score, the pale ones have lost theirs.
DDOST in a dopamine neuron, astrocyte and microglia tri-culture
0.82
Knock down and overexpress DDOST in a dopamine neuron, astrocyte and microglia tri-culture; measure LRRK2 kinase activity (pT73-Rab10) and lysosome position and dopamine neuron survival after α-synuclein fibril challenge.
Knock down and overexpress SH3GL2 in dopamine neurons; measure LRRK2 kinase activity (pT73-Rab10) and lysosome position and survival of TH-positive neurons and phospho-α-synuclein (pS129) inclusions after seeding with α-synuclein fibrils.
genetics
0.88
cell type
1.00
chemistry
0.45
trials
1.00
literature
0.61
pathways
1.00
no reader dissentedSH3 domain containing GRB2 like 2, endophilin A1
Knock down and overexpress DNAJC6 in oligodendrocytes; measure LRRK2 kinase activity (pT73-Rab10) and lysosome position and α-synuclein uptake and myelin basic protein output under oxidative stress.
genetics
0.83
cell type
1.00
chemistry
0.27
trials
1.00
literature
0.42
pathways
1.00
dissent: chemistry, literatureDnaJ heat shock protein family (Hsp40) member C6
Knock down and overexpress ITPKB in astrocytes; measure a transcriptome-wide state shift (bulk RNA-seq) and α-synuclein uptake and degradation, and dopamine neuron survival in co-culture.
genetics
0.82
cell type
0.75
chemistry
0.60
trials
1.00
literature
0.67
pathways
0.30
dissent: pathwaysinositol-trisphosphate 3-kinase B
Knock down and overexpress RIT2 in dopamine neurons; measure a transcriptome-wide state shift (bulk RNA-seq) and survival of TH-positive neurons and phospho-α-synuclein (pS129) inclusions after seeding with α-synuclein fibrils.
genetics
0.82
cell type
1.00
chemistry
0.45
trials
1.00
literature
0.51
pathways
0.30
dissent: pathwaysRas like without CAAX 2
GBA1 in a dopamine neuron, astrocyte and microglia tri-culture
0.69
Knock down and overexpress GBA1 in a dopamine neuron, astrocyte and microglia tri-culture; measure glucocerebrosidase (GCase) activity and LC3-II and p62 flux and dopamine neuron survival after α-synuclein fibril challenge.
PLA2G6 in a dopamine neuron, astrocyte and microglia tri-culture
0.69
Knock down and overexpress PLA2G6 in a dopamine neuron, astrocyte and microglia tri-culture; measure LRRK2 kinase activity (pT73-Rab10) and lysosome position and dopamine neuron survival after α-synuclein fibril challenge.
genetics
0.88
cell type
0.40
chemistry
0.60
trials
1.00
literature
0.29
pathways
1.00
dissent: cell type, literaturephospholipase A2 group VI
GAK in a dopamine neuron, astrocyte and microglia tri-culture
0.68
Knock down and overexpress GAK in a dopamine neuron, astrocyte and microglia tri-culture; measure LRRK2 kinase activity (pT73-Rab10) and lysosome position and dopamine neuron survival after α-synuclein fibril challenge.
genetics
0.76
cell type
0.40
chemistry
0.60
trials
1.00
literature
0.46
pathways
1.00
dissent: cell typecyclin G associated kinase
PSAP in a dopamine neuron, astrocyte and microglia tri-culture
0.68
Knock down and overexpress PSAP in a dopamine neuron, astrocyte and microglia tri-culture; measure glucocerebrosidase (GCase) activity and LC3-II and p62 flux and dopamine neuron survival after α-synuclein fibril challenge.
Knock down and overexpress GRIN2C in astrocytes; measure evoked dopamine release and dopamine transporter uptake and α-synuclein uptake and degradation, and dopamine neuron survival in co-culture.
genetics
0.00
cell type
1.00
chemistry
1.00
trials
1.00
literature
0.82
pathways
1.00
dissent: geneticsglutamate ionotropic receptor NMDA type subunit 2C
Knock down and overexpress GRIN2D in oligodendrocytes; measure evoked dopamine release and dopamine transporter uptake and α-synuclein uptake and myelin basic protein output under oxidative stress.
genetics
0.00
cell type
1.00
chemistry
1.00
trials
1.00
literature
0.82
pathways
1.00
dissent: geneticsglutamate ionotropic receptor NMDA type subunit 2D
EIF4G1 in a dopamine neuron, astrocyte and microglia tri-culture
0.66
Knock down and overexpress EIF4G1 in a dopamine neuron, astrocyte and microglia tri-culture; measure glucocerebrosidase (GCase) activity and LC3-II and p62 flux and dopamine neuron survival after α-synuclein fibril challenge.
A receipt is printed when a translation score is below 0.55, a gene is not found, a source fails mid-run, a finding can't be recovered from the combined vector, an atlas prediction does not replicate, or a sealed study misses its rule. Each receipt contains the hash of the one before it. Your browser recomputes the hashes.
loading the receipts
Swarm
A failed prediction is not the end of the experiment. When an atlas prediction fails, six analysts open a case on its receipt, each from a different angle:
power: is the effect absent, or real but too small for 9 donors per half?
confounder: does it survive adjustment for age, sex and post-mortem interval?
composition: is it a change within cells, or a shift in the mix of cells as the dopamine neurons die?
cell type: is the gene's effect stronger in another cell group?
axis: does the gene track another pathology measure better?
skeptic: does the result rest on one donor, or on the most severely affected donors?
The cell-type and axis analysts propose new predictions. The confounder, composition and skeptic analysts challenge every proposal, and only proposals that survive all three go forward, sealed and tested on tissue the agent has not read: the substantia nigra, an independent cohort, at a significance level divided by the number of follow-ups so far. The gym scores how well these analysts judge.
6,476 cases, 267 follow-ups sealed, 172 pointing the same way in the second region, 11 confirmed.
full sample -0.01 (p 0.976, n 18), 95% CI -0.47 to +0.46. the interval is wide: not enough donors to tell.
confounder
adjusted for sex: +0.04 (p 0.881, n 18). does not hold up.
composition
dopamine neurons share of nuclei vs axis +0.05 (p 0.854); controlling for it: -0.02 (p 0.949, n 18). could be the mix.
skeptic
leave-one-out range -0.15 to +0.13, sign flips 8; without severely affected donors -0.01 (p 0.976, n 18). fragile.
cell type
no other cell group does better.
axis
no measure beats "tracks Braak Lewy-body stage" for LAT in dopamine neurons.
case closed: nothing better to follow up.
Sealed studies
Between genes the agent runs studies on whole sets of Parkinson's genes: the lysosomal genes, the dopamine identity programme, mitochondrial quality control. Each is written down with its rule and hashed before its data is read, run once, posted to the live feed, and prints a receipt if it misses. The lysosome program is the first.
Before a prediction is tested, its hash can go on-chain. Predictions you sign with a wallet are stored in the registry contract on Robinhood Chain, an Ethereum L2; the agent pays the gas, writes the split-half result back, and a prediction that replicates can earn the wallet that made it a reward. Holders can put a gene at the front of the agent's queue. The registry contract is live.
Parkinson, J. An Essay on the Shaking Palsy. London: Sherwood, Neely and Jones (1817). Reprinted in J Neuropsychiatry Clin Neurosci 14, 223–236 (2002). doi:10.1176/jnp.14.2.223
Volpato, D. et al. A deep cellular atlas of the human ventral substantia nigra in Parkinson's identifies a genetic and molecular overlap with insulin resistance. medRxiv (2025), preprint; data GEO GSE329625. www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE329625
Kamath, T. et al. Single-cell genomic profiling of human dopamine neurons identifies a population that selectively degenerates in Parkinson's disease. Nature Neuroscience 25, 588–595 (2022). doi:10.1038/s41593-022-01061-1
Martirosyan, A. et al. Unravelling cell type-specific responses to Parkinson's disease at single cell resolution. Molecular Neurodegeneration 19, 7 (2024). doi:10.1186/s13024-023-00699-0
Wang, Q. et al. Molecular profiling of human substantia nigra identifies diverse neuron types associated with vulnerability in Parkinson's disease. Science Advances 10, eadi8287 (2024). doi:10.1126/sciadv.adi8287
Aligning Science Across Parkinson's (ASAP) Collaborative Research Network Cloud: harmonised post-mortem brain single-nucleus data. cloud.parkinsonsroadmap.org
Braak, H. et al. Staging of brain pathology related to sporadic Parkinson's disease. Neurobiology of Aging 24, 197–211 (2003). doi:10.1016/S0197-4580(02)00065-9
Spillantini, M. G. et al. α-Synuclein in Lewy bodies. Nature 388, 839–840 (1997). doi:10.1038/42166
Nalls, M. A. et al. Identification of novel risk loci, causal insights, and heritable risk for Parkinson's disease: a meta-analysis of genome-wide association studies. Lancet Neurology 18, 1091–1102 (2019). doi:10.1016/S1474-4422(19)30320-5
Kim, J. J. et al. Multi-ancestry genome-wide association meta-analysis of Parkinson's disease. Nature Genetics 56, 27–36 (2024). doi:10.1038/s41588-023-01584-8
Sidransky, E. et al. Multicenter analysis of glucocerebrosidase mutations in Parkinson's disease. New England Journal of Medicine 361, 1651–1661 (2009). doi:10.1056/NEJMoa0901281
Cheng, H.-C., Ulane, C. M. & Burke, R. E. Clinical progression in Parkinson disease and the neurobiology of axons. Annals of Neurology 67, 715–725 (2010). doi:10.1002/ana.21995
Nelson, M. R. et al. The support of human genetic evidence for approved drug indications. Nature Genetics 47, 856–860 (2015). doi:10.1038/ng.3314
Plate, T. A. Holographic reduced representations. IEEE Transactions on Neural Networks 6, 623–641 (1995). doi:10.1109/72.377968
.jpg){kind=link}
.png){kind=link}
