Probabilistic Modelling

FA-DPS: A Composite Severity Score for Friedreich Ataxia

Developing a Composite Severity Score for Friedreich’s Ataxia: A Robust Alternative to mFARS for Predictive Modeling

A better target for machine learning: a biologically grounded disease-progression score instead of a subjective clinical rating.

The problem

Sensor models are trained to predict a clinical scale that is subjective, nonlinear and limited in sensitivity, so it makes a weak ground truth.

The approach

FA-DPS models clinical and sensor biomarkers as sigmoidal trajectories along a latent disease timeline, estimating patient-specific progression and shared biomarker dynamics.

The outcome

Models trained on FA-DPS predicted more accurately than models trained on the clinical scale across all three instrumented devices.

In detail

Developed FA-DPS, a composite disease-progression score that models clinical and sensor-derived biomarkers along a latent disease timeline. The framework uses biologically plausible sigmoidal trajectories to capture shared progression dynamics and patient-specific variation. It was designed as a more robust and interpretable target for machine-learning models than relying solely on mFARS.

Built with

Biomedical AI
Disease Progression
Latent Variable Modelling