Variability vs Phenotype: multimodal analysis of Dravet Syndrome Brain Organoids powered by Deep Learning

Abstract

Brain organoids (BO) have risen as a reliable model for neurodelopmental disorders (ND), reproducing human brain development milestones. However, their significant intra- and inter-organoid variability compromises their use in advanced tasks such as drug testing. Overcoming experimental variability is crucial for models prone to variation, like unguided BO. BO modelling in Dravet Syndrome, a late-onset epileptic ND, represents a great challenge since BO variability accumulates with time, when phenotype shows in vitro. Leveraging deep learning, we developed ImPheNet, a predictive tool grounded in BO live imaging datasets. ImPheNet accurately classified phenotypes and assessed drug toxicity in BO derived from DS, revealing differences between genotypes and upon antiseizure drug exposure. These results are supported by transcriptomic and functional data, revealing an excitatory-inhibitory imbalance during the maturation of DS organoids. Altogether, our DL-predictive live imaging strategy, ImPheNet, emerges as a powerful tool enhancing BO research and advancing ND treatments.