Field: Medicine › NeurologyIllustrative exampleArea: alsTarget: C9orf72 repeat RNAStage: phase 1✦ AI inferred reasoning
Hypothesis
Intrathecal antisense against the C9orf72 repeat expansion would lower dipeptide repeat protein throughout the CNS.
What failed
Poly(GP) fell sharply in lumbar CSF but post-mortem cortex from two participants showed near-baseline levels.
Why
Insufficient exposure above the cervical cord — intrathecal delivery did not achieve tissue distribution to the motor cortex, so the hypothesis was never actually tested there.
Field: Medicine › Endocrinology, Diabetes and MetabolismIllustrative exampleArea: type 2 diabetesTarget: GPR40 agonismStage: preclinical✦ AI inferred reasoning
Hypothesis
GPR40 agonism would boost glucose-stimulated insulin secretion without the hypoglycemia risk of sulfonylureas.
What failed
Efficacy was strong in rodent models but did not replicate in primate islet studies.
Why
Species-specific receptor expression patterns — rodent islets over-express GPR40 relative to primate and human islets.
Field: Medicine › RheumatologyIllustrative exampleArea: rheumatoid arthritisTarget: JAK1Stage: preclinical✦ AI inferred reasoning
Hypothesis
A JAK1-selective inhibitor would retain efficacy in collagen-induced arthritis while avoiding pan-JAK haematologic effects.
What failed
Efficacy held in rat models but the therapeutic window closed in dog toxicology before an efficacious exposure was reached.
Why
Species differences in JAK1 dependence made the rodent window misleading; the programme stopped on narrow therapeutic index rather than on any efficacy finding.
A TREM2 agonist antibody would boost microglial plaque clearance and slow decline in early Alzheimer's.
What failed
Microglial activation markers rose and soluble TREM2 in CSF dropped, but CDR-SB scores were unchanged at 12 months.
Why
Microglial engagement was achieved without a downstream effect on the neurodegenerative cascade — activation alone does not appear to be rate-limiting at this stage.