Exploring Membrane-Damaging Effects of Amyloid Aggregates on the Clustering Behaviors of Lipid Molecules in Simulated Neuronal Membranes Using Unsupervised Machine Learning (NMFk) Algorithm

Recent experimental studies have indicated that neuronal membranes contain dynamic, phase-separated nanodomains distributed on both leaflets of the lipid bilayers. Using microsecond MD simulations, we have created many phase-separated lipid bilayers containing ordered (Lo), disordered (Ld), mixed Lo and Ld (Lod), and ganglioside-cluster (GM1) domains that mimic the neuronal membranes. We have further demonstrated Lod and GM1 domains represent major membrane damage targets of several cytotoxic amyloid protein oligomers associated with Alzheimer’s disease (AZ). At present, a robust computational tool to characterize the membrane disruptive effects of these oligomers based on the clustering behaviors of nanodomains is not available. By implementing the Non-Negative Matrix Factorization (NMFk), an unsupervised Machine Learning (ML) algorithm, we have decomposed the time-dependent lipid-lipid proximity matrix into two latent signatures and activities matrices, sorted all the constituent lipids into domains, and assessed the clustering disruption behaviors of nanodomains through the transient change in the signatures assignment of each lipid. The results of this ML study will provide new insights into the membrane-damaging mechanisms of cytotoxic amyloid oligomers linked to the early pathogenesis of AZ.