Diagnosed Decades Too Late

Alzheimer's, Parkinson's, ALS, and other neurodegenerative diseases share a common molecular mechanism: intrinsically disordered proteins (IDPs) misfold and aggregate into toxic oligomers and amyloid fibrils that kill neurons. By the time clinical symptoms appear (memory loss, tremor, motor weakness), 50–80% of the vulnerable neurons are already dead. No diagnostic technology can reliably detect the earliest stages of protein aggregation — the point where intervention could prevent neuronal death rather than merely slow decline. The fundamental barrier is that IDPs are inherently flexible, adopting ensembles of transient conformations rather than stable structures, making them invisible to conventional structural biology tools (X-ray crystallography, cryo-EM) and resistant to computational prediction.

Neurodegenerative diseases affect over 55 million people worldwide (Alzheimer's alone: 55 million; Parkinson's: 10 million). Global costs exceed $1 trillion annually. Every major Phase 3 clinical trial for Alzheimer's disease that has targeted amyloid plaques (the end-stage aggregation product) has either failed or shown only marginal benefit, leading to the hypothesis that intervention must occur much earlier — at the initial misfolding/oligomerization stage. But this pre-symptomatic stage cannot currently be detected, creating a catch-22: treatments must be given before diagnosis is possible.

Amyloid PET scans can detect amyloid plaque burden but only after substantial aggregation has occurred — they cannot detect the early oligomeric species believed to be most toxic. Blood-based biomarkers (phospho-tau, neurofilament light chain, amyloid-beta 42/40 ratio) correlate with disease progression but measure downstream consequences of aggregation, not the aggregation process itself. Computational prediction of IDP behavior is fundamentally limited because these proteins do not adopt a single structure — they exist as dynamic ensembles of thousands of conformations. AlphaFold and other AI protein structure tools were designed for folded proteins and perform poorly on IDPs. Therapeutic antibodies targeting specific aggregation states (e.g., aducanumab, lecanemab) have shown that clearing plaques produces modest clinical benefit, reinforcing the view that the therapeutic window is earlier than current diagnostics can reach.

Two advances are needed: (1) AI/ML models trained on molecular dynamics data that can predict which IDP conformational states lead to aggregation and identify the earliest detectable markers of misfolding — essentially a "grammar" of protein aggregation that maps sequence features to aggregation propensity; (2) ultrasensitive biosensors that can detect early-stage aggregation intermediates (oligomers, protofibrils) in accessible biofluids (blood, CSF) before they accumulate to pathological levels. Together, these would enable pre-symptomatic screening and define the therapeutic window for aggregation-preventing interventions.

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.