Artificial neural network applied to fragile X-associated tremor/ataxia syndrome stage diagnosis based on peripheral mitochondrial bioenergetics and brain imaging outcomes

Abstract

Abstract No proven prognosis is available for the neurodegenerative disorder fragile X-associated tremor/ataxia syndrome (FXTAS). Artificial neural network analyses (ANN) were used to predict FXTAS progression using data from 127 adults (noncarriers and FMR1 premutation carriers with and without FXTAS) with five outcomes from brain MRI imaging and 22 peripheral bioenergetic outcomes from two cell types. Diagnosis accuracy by ANN predictions ranged from 41.7 to 86.3% (depending on the algorithm used), and those misclassified usually presented a higher FXTAS stage. ANN prediction of FXTAS stages was based on a combination of two imaging findings (white matter hyperintensity and whole-brain volumes adjusted for intracranial volume) and four bioenergetic outcomes. Those at Stage 3 vs. 0–2 showed lower mitochondrial mass, higher oxidative stress, and an altered electron transfer consistent with mitochondrial unfolded protein response activation. Those at Stages 4–5 vs. 3 had higher oxidative stress and glycerol-3-phosphate-linked ATP production, suggesting that targeting mGPDH activity may prevent a worse prognosis. This was confirmed by the bioenergetic improvement of inhibiting mGPDH with metformin in affected fibroblasts. ANN supports the prospect of an unbiased molecular definition in diagnosing FXTAS stages while identifying potential targets for personalized medicine.