Title : Decoding human neural progenitor fate commitment with deep learning: A single-cell transcriptomic framework for predicting developmental cell-state transitions
Abstract:
Focal Cortical Dysplasia (FCD) causes epilepsy in 4 million newborns yearly. Treatments are surgical removal of malformed brain tissue, which fails for 40% of patients and causes permanent cognitive defects. FCD arises when neural progenitor cells (NPCs) commit to regional brain identities at incorrect embryonic developmental time points. However, this commitment in humans has never been precisely identified, preventing implementation of regenerative therapies. This project aimed to identify the precise time point at which NPC fate commitment occurs to guide regenerative therapy for FCD. Using single-cell RNA sequencing from 52,183 human brain organoid cells across two independent datasets during organoid development, I calculated fate-bias scores (rostral gene markers - caudal gene markers) to track forebrain and hindbrain regions. Interquartile-range-based heterogeneity analysis showed a fall in transcriptional variability in correspondence with the commitment threshold. Commitment occurred 18±2 days of brain development across all datasets, marking it the first measurement of this threshold in humans. Cells showed reduced plasticity and became irreversibly committed to their regional fate after this timepoint. A gradient boosting machine learning model achieved 93% accuracy in predicting if cells would commit at normal timing using Day 10-14 gene expression profiles. Defining this enabled ML prediction from Days 10-14, flagging abnormal commitment batches for early chemical intervention while cells were still plastic. This work identified the first quantitative time-point of neural commitment in human development, the intervention window before cells irreversibly lock into rostro-caudal fates. Enabling the manufacturing of regenerative cortical tissue that eliminates seizures while preserving cognitive function.
