Extended insight into selective vulnerability of neurons in AD using machine learning approach

Abstract Background Alzheimer’s disease is characterized by the spread of tau pathology and loss of neurons with a specific spatio‐temporal pattern. Recent publications have attempted to uncover the identity of the selectively vulnerable neurons using single‐nucleus transcriptomic data derived from affected brain regions from patients with Alzheimer’s disease. Correlating advancing tau pathology as measured by Braak staging with neuronal loss, one of the major markers reported for the susceptible excitatory neurons was RORB, which was also validated using quantitative neuropathological methods. However, the heterogeneity of RORB expression in the highlighted clusters suggested that not all RORB expressing neurons may be susceptible. Besides, further classifying neuronal population into sub‐populations shows population specific RORB expression. Method Therefore, we further extended our insight into selective vulnerability of neuronal sub‐populations using a machine learning (ML) approach. We identified a transcriptionally homogeneous population using ML‐based classifiers with high accuracy and AUC > 0.90. Result Our analysis shows that fine mapping of transcriptionally similar neurons is key to pinpoint sub‐populations that are selectively lost during later Braak stages. For instance, refined sub‐population analysis revealed that neuronal populations with moderate RORB expression were lost while those with higher RORB expression do not show significant change with advancing Braak stage. Conclusion Our analysis provides a reference of neuronal sub‐populations and associated markers that could be further interrogated to uncover potential mechanisms that underlie selective degeneration in Alzheimer’s Disease.