health · digital · family: it worked in the lab
diagnosed decadestoo late
Neurodegenerative diseases are diagnosed decades after protein aggregation begins — when neuronal damage is irreversible
Problem statement
Alzheimer's, Parkinson's, ALS, and other neurodegenerative diseases share a common molecular mechanism: flexible proteins, including intrinsically disordered proteins (IDPs), misfold and stick together into clumps (aggregates) that damage cells. (Parkinson's disease is a synucleinopathy: the protein alpha-synuclein is a major component of the Lewy bodies that mark it; Cheng et al. 2010, Kordower et al. 2013.) Most treatments target the large clumps that appear late in disease, when much of the damage is already done and it is very hard to restore normal function. In Parkinson's disease, a review of post-mortem and imaging studies estimates that by the time motor symptoms appear, about 30% of the dopamine neurons of the substantia nigra and 50–60% of their axon terminals in the striatum have already been lost (Cheng et al. 2010); a later post-mortem series suggests at least a 50% loss of dopamine-producing fibres in the putamen at symptom onset and found near-complete loss in the dorsal putamen by four years after diagnosis (Kordower et al. 2013). In one study of families with inherited Alzheimer's disease, a fall in amyloid-beta in the spinal fluid was detected about 25 years before the expected onset of symptoms, and amyloid deposits in the brain about 15 years before (Bateman et al. 2012). Scientists do not have good tools to predict, prevent, or reverse the early misfolding and clumping events, and ARPA-H is asking for technologies that can detect their earliest signs. That early stage is the point where intervention could stop the disease before it develops, rather than only treating symptoms after serious damage has occurred. A major challenge is that IDPs are inherently flexible: they do not fold into stable, well-defined three-dimensional structures but fluctuate rapidly over an ensemble of conformations, and very few full-length IDPs have been characterized structurally (Wright and Dyson 2015). Existing drugs cannot capture or influence them, and many are considered "undruggable."
Why this matters
ARPA-H's director calls the fight against neurodegenerative disease a century-long battle. The World Health Organization reports that in 2021, 57 million people were living with dementia worldwide, that Alzheimer's disease may account for 60–70% of dementia cases, and that in 2019 dementia cost economies globally US$1.3 trillion (WHO dementia fact sheet, 2026); WHO's global estimate for 2019 was over 8.5 million people with Parkinson's disease (WHO Parkinson disease fact sheet, 2023). Protein misfolding and clumping can lead to Alzheimer's disease, ALS, some cancers, and other chronic and degenerative conditions. Most current treatments target the large protein clumps that appear late in disease. A 2024 systematic review and meta-analysis of randomized trials of eight anti-amyloid monoclonal antibodies (19 publications, 23,202 participants) found small improvements over placebo on cognitive and functional scales, none of which, including those for lecanemab, aducanumab, and donanemab, exceeded the minimal clinically important difference (Ebell et al. 2024). ARPA-H's position is that intervention must occur much earlier, at the very beginning of the misfolding process. But there are no good tools to predict or detect that early stage, creating a catch-22: treatments must be given before diagnosis is possible.
What’s been tried and why it hasn’t worked
Most current treatments target the large protein clumps that appear late in disease. By that point much of the damage is done, and it is very hard to restore normal function. Existing drugs cannot capture or influence the flexible proteins involved, so many of these proteins are considered "undruggable." The antibodies lecanemab and aducanumab were studied in large randomized trials that found substantial reductions in amyloid deposition but only modest improvements in cognition and function (Ebell et al. 2024). Scientists do not have good tools to predict, prevent, or reverse the early misfolding and clumping events. ARPA-H also describes the research on flexible, disordered proteins as divided into silos.
What would unlock progress
ARPA-H's program names two technical focus areas: (1) next-generation AI and experimental models for IDPs and aggregation, to decode the "grammar" that guides how flexible, disordered proteins behave; (2) new kinds of therapy that interfere with or prevent IDP misfolding and aggregation while preserving the proteins' critical functions, together with new detection capabilities that respond to early-stage IDP aggregation. Together, these would make it possible to find and fix problems with proteins before they clump together and damage cells.
Entry points for student teams
A student team could train machine learning models on existing molecular dynamics simulation datasets of known amyloidogenic peptides (e.g., amyloid-beta, alpha-synuclein) to predict aggregation-prone conformational states from sequence features alone. A biosensor-oriented team could develop and characterize a surface plasmon resonance or nanopore-based assay for detecting sub-nanomolar concentrations of amyloid-beta oligomers in spiked buffer, benchmarking against current ELISA sensitivity. Relevant disciplines: computational biology, biophysics, biomedical engineering, machine learning.
Genes
Structural cousins: stuck the same way, in other fields
Sources
ARPA-H, "BIOmolecular Grammar for protein Aggregation Modulation and Intervention (BIOGAMI)," ARPA-H press release, "Rewriting protein 'grammar' to stop neurodegenerative disease before it starts," January 30, 2026, accessed 2026-02-23; Bateman, R.J., Xiong, C., Benzinger, T.L.S., et al., "Clinical and biomarker changes in dominantly inherited Alzheimer's disease," N Engl J Med, 367(9):795-804, 2012, accessed 2026-09-29; World Health Organization, "Dementia," fact sheet, 3 July 2026, accessed 2026-09-29; World Health Organization, "Parkinson disease," fact sheet, 9 August 2023, accessed 2026-09-29; Cheng, H.C., Ulane, C.M., Burke, R.E., "Clinical progression in Parkinson disease and the neurobiology of axons," Ann Neurol, 67(6):715-725, 2010, accessed 2026-09-29; Kordower, J.H., Olanow, C.W., Dodiya, H.B., Chu, Y., Beach, T.G., Adler, C.H., Halliday, G.M., Bartus, R.T., "Disease duration and the integrity of the nigrostriatal system in Parkinson's disease," Brain, 136(8):2419-2431, 2013, accessed 2026-09-29; Ebell, M.H., Barry, H.C., Baduni, K., Grasso, G., "Clinically Important Benefits and Harms of Monoclonal Antibodies Targeting Amyloid for the Treatment of Alzheimer Disease: A Systematic Review and Meta-Analysis," Ann Fam Med, 22(1):50-62, 2024, accessed 2026-09-29; Wright, P.E., Dyson, H.J., "Intrinsically disordered proteins in cellular signalling and regulation," Nat Rev Mol Cell Biol, 16(1):18-29, 2015, accessed 2026-09-29 go to source 1 ↗ go to source 2 ↗ go to source 3 ↗ go to source 4 ↗ go to source 5 ↗ go to source 6 ↗ go to source 7 ↗ go to source 8 ↗ go to source 9 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
Related briefs: `health-tbi-biomarker-clinical-adoption` (biomarker-to-clinical-action translation challenges — relevant if early aggregation markers are discovered); `bio-genotype-phenotype-prediction-gap` (similar pattern of needing to predict complex biological outcomes from sequence data). The `failure:wrong-problem` tag reflects the decades of therapeutic effort directed at clearing end-stage plaques rather than preventing early aggregation — the field targeted the wrong stage of the disease process. The brief is tagged `temporal:static`: the early-detection capability gap itself is persistent rather than accelerating. (Aging populations do mean neurodegenerative disease prevalence is rising, but that is growth in downstream burden, not measurable worsening of the detection problem in the sense the `temporal:worsening` test requires.) Source-bias note: ARPA-H frames AI protein modeling as newly tractable; while ML tools for IDPs are advancing, predicting the full conformational ensemble of disordered proteins remains at the frontier of computational biophysics.
Reconciliation 2026-09-29: Fact repair after a site review read this write-up against its two cited sources. Both were fetched in full on 2026-09-29: the ARPA-H BIOGAMI program page (https://arpa-h.gov/explore-funding/programs/biogami) and the ARPA-H announcement linked from it (https://arpa-h.gov/news-and-events/rewriting-protein-grammar-stop-neurodegenerative-disease-it-starts). Many figures and names in the write-up were not in either page, and no other source was cited for them. They were removed or replaced with what the pages say, and were not re-sourced. (1) The Source line dated the announcement 2025. The page gives January 30, 2026. The Source line now carries that date and the address of the announcement. (2) The text said that by the time symptoms appear "50–80% of the vulnerable neurons are already dead." ARPA-H gives no figure. It says most treatments target large clumps that show up late in disease, when much of the damage is already done. The text now says that. (3) Three figures in "Why This Matters" are not in either page and were removed: "over 55 million people worldwide", "Parkinson's: 10 million", and global costs above "$1 trillion annually." (4) The text said every major Phase 3 Alzheimer's trial aimed at amyloid plaques "has either failed or shown only marginal benefit." ARPA-H does not mention trials. The sentence now gives ARPA-H's position that treatment has to start at the very beginning of the misfolding process. (5) "What's Been Tried" named amyloid PET scans, three blood biomarkers, AlphaFold, and the antibodies aducanumab and lecanemab. None of these is named in either page. The section now reports what ARPA-H says about current approaches: treatments target late-stage clumps, existing drugs cannot capture the flexible proteins, there are no good tools for the early events, and the research is divided into silos. (6) The Problem Statement named Parkinson's disease, "toxic oligomers and amyloid fibrils," X-ray crystallography and cryo-EM. The pages name Alzheimer's disease, ALS and some cancers, speak of clumps that damage cells, and say the flexible proteins are hard for drugs to target. The paragraph now follows the pages. (7) "What Would Unlock Progress" described two advances, models trained on molecular dynamics data and biosensors for blood and spinal fluid. The program page names two technical focus areas in different terms, and the second one includes new therapies as well as detection. The section now gives the two focus areas as the page states them. Confirmed against the pages and left as written: the program name and its spelling in the Source line; the title of the announcement; intrinsically disordered proteins as the subject; the "grammar" framing; the use of AI and machine learning. NOT changed, and flagged for a decision: the word "Decades" in the title and the short title, and the word "Irreversible" in the title. Neither page gives a time span between the start of aggregation and diagnosis, and the pages say restoring normal function is "very hard" at the late stage, which is weaker than irreversible. The titles were left alone because nothing in the pages contradicts them outright. NOT checked: the suggestions in the entry-points section, which are advice and do not report the source, including whether the simulation datasets it mentions are public; the phrase "decades of therapeutic effort" in the tag rationale above; and the full solicitation on SAM.gov (ARPA-H-SOL-26-149), which the program page links to but this write-up does not cite, and which was not read. Later the same day the editor asked whether "decades" in the titles could be sourced. Bateman et al. 2012, read in full from the PubMed Central copy on 2026-09-29, report in their abstract that "Concentrations of amyloid-beta (Aβ)42 in the CSF appeared to decline 25 years before expected symptom onset" and that amyloid deposition "was detected 15 years before expected symptom onset" in carriers of dominantly inherited Alzheimer's mutations. That paper was added to the Source line and one sentence citing it was added to the Problem Statement. The titles keep "decades"; the word "irreversible" in the main title rests on the ARPA-H page's statement that it is very hard to get cells working normally again, and was left as written. Genes untouched.
Reconciliation 2026-09-29 (re-sourcing): The editor asked for the claims removed in round 5 to be re-sourced, with each one put back only where a primary source states it. Each source below was downloaded and read on 2026-09-29 at the address given. (1) "Why This Matters" now gives the World Health Organization's figures instead of the removed ones: 57 million people living with dementia worldwide in 2021, Alzheimer's disease accounting for 60–70% of dementia cases, and a global cost of dementia of US$1.3 trillion in 2019, all from the WHO dementia fact sheet dated 3 July 2026 (https://www.who.int/news-room/fact-sheets/detail/dementia); and over 8.5 million people with Parkinson's disease in 2019, from the WHO Parkinson disease fact sheet dated 9 August 2023 (https://www.who.int/news-room/fact-sheets/detail/parkinson-disease). These differ from the original text, which said 55 million for Alzheimer's alone, 10 million for Parkinson's, and costs above $1 trillion: the 57 million is for all dementia, and the Parkinson's estimate is 8.5 million. (2) The removed sentence that 50–80% of vulnerable neurons are dead by the time symptoms appear was not restored as written. The two Parkinson's papers read give different figures: Cheng, Ulane and Burke 2010 (https://pmc.ncbi.nlm.nih.gov/articles/PMC2918373/) conclude that at first diagnosis "only 30% or so of dopamine neurons of the SN have been lost and only 50-60% of their axon terminals," and Kordower et al. 2013 (https://pmc.ncbi.nlm.nih.gov/articles/PMC3722357/) report "at symptom onset, there is a moderate loss of at least 50% of fibres in the putamen" and that loss of dopaminergic markers in the dorsal putamen "is virtually complete by 4 years post-diagnosis." The Problem Statement now gives those figures for Parkinson's disease only, worded as the papers' estimates rather than as direct post-mortem counts: Cheng et al. combine post-mortem and in-vivo imaging data for the terminal figure, and Kordower et al. infer the loss at onset from brains examined one year or more after diagnosis (an independent check on 2026-09-29 tightened this wording). No source was found that gives a single figure across all neurodegenerative diseases. (3) Parkinson's disease is back in the list of diseases in the Problem Statement, on the basis of Kordower et al. 2013 ("Parkinson's disease is a synucleinopathy") and Cheng et al. 2010 ("α-synuclein protein is a major component of Lewy bodies"). The statement that IDPs adopt ensembles of conformations rather than stable structures is back, resting on Wright and Dyson 2015 (https://pmc.ncbi.nlm.nih.gov/articles/PMC4405151/), which says such proteins "fluctuate rapidly over an ensemble of conformations" and that "very few full-length IDPs have been characterized structurally." The words "toxic oligomers and amyloid fibrils" and the claim that IDPs are invisible to X-ray crystallography and cryo-EM stay out: Wright and Dyson say bound-state structures of disordered regions can be obtained by X-ray crystallography "if well ordered," which is a narrower statement, and the α-synuclein review sought for oligomers and fibrils (Lashuel et al. 2013) could not be read because the PubMed Central copy returned a browser check. (4) The sentence that every major Phase 3 anti-amyloid trial has failed or shown marginal benefit was replaced by what a systematic review states. Ebell et al. 2024 (https://pmc.ncbi.nlm.nih.gov/articles/PMC11233076/) pooled 19 publications with 23,202 participants across eight anti-amyloid antibodies and found small improvements over placebo, and "None of the changes, including those for lecanemab, aducanumab, and donanemab, exceeded the MCID." The same paper supports the sentence in "What's Been Tried" that lecanemab and aducanumab trials "found substantial reductions in amyloid deposition but only modest improvements in cognition and function." The original's claim that these results led to the hypothesis that treatment must start earlier is not in the paper and stays out; the text keeps ARPA-H's own position on timing. (5) Amyloid PET, the three blood biomarkers, and AlphaFold stay out of "What's Been Tried." The biomarker review sought (Hansson 2021, Nat Med) has no open copy; the AlphaFold paper sought (Ruff and Pappu 2021, J Mol Biol) is behind the publisher's block, and the open paper found instead (Alderson et al. 2023, PNAS) says that thousands of disordered regions receive confident AlphaFold2 predictions, which is a different statement from the original's "performs poorly on IDPs." (6) The original "What Would Unlock Progress" text, which named models trained on molecular dynamics data and biosensors for oligomers in blood and spinal fluid as the two needed advances, is a recommendation rather than a reported fact and was not re-sourced; the section keeps the two focus areas as ARPA-H states them. Genes untouched.