Towards AI-driven prediction of HTT CAG size in super-expanded human spiny projection neurons from Huntington disease donors.

Somatic instability (SI) of the CAG tract in HTT is a major driver of neurodegeneration of Spiny Projection Neurons (SPNs), the primary neuronal subtype affected in Huntington's disease (HD). SPNs can accumulate hundreds of CAG repeats during a patient's lifetime, and once the expansion exceeds ∼150 CAGs, they acquire distinct, cell-autonomous transcriptional alterations that ultimately contribute to degeneration. Here, we developed the "HD-Phase-Model", a mathematical framework designed to identify "super-expanded" SPNs without repeat sizing, by leveraging the only available single-nucleus HD post-mortem dataset that provides both transcriptional profile and matched HTT CAG sizes. After validating model performance on the test data, we applied it to independent single-nucleus datasets lacking CAG sizing information and across multiple brain regions. In all cases, the model consistently detected SPNs populations with convergent transcriptional dysregulation signatures indicative of extreme CAG expansion.Importantly, although the model was trained on caudate SPNs, we observed highly similar dysregulation patterns in putamen and accumbens, while no evidence of super-expansion was found in SPNs from Alzheimer's and Parkinson's disease donors.Together, these findings demonstrate that transcriptional profiles alone can serve as predictors of HTT CAG size, enabling systematic identification of super-expanded SPNs and providing insights into HD-specific neurodegenerative mechanisms.