BPS2026 – Applications of unsupervised machine learning for clustering-based analysis of tau dimer residue contact maps across multi-target compound treatments

Tau protein aggregation is a hallmark of Alzheimer's disease (AD) and related tauopathies. Despite its central role in disease progression, few treatments directly target the generation of misfolded tau and its aggregation in the early pathogenesis of AD. To advance tau-directed therapeutics, it is critical to understand how tau dimers, the smallest tau aggregates, respond to multi-targeted compound (MTC) interventions. However, no standardized, objective methods currently exist to analyze MTC-induced disruptions of interchain protein residue contact maps from all-atom molecular dynamics (MD) simulations. This project introduces an unsupervised machine-learning pipeline to analyze residue contact map data derived from our recent post-docking MD simulations of various MTC-tau dimer complexes in solution and on raft-like membrane surfaces. Our MTCs contain two HDAC-6 inhibitors (L17 and L20) and two chemical analogs (C25 and C44) of CNS-11, a known tau-fibril inhibitor. Residue contact maps, representing inter-residue distances among the residues of the 130-residue-long tau-k18 monomers, were converted into high-dimensional feature vectors and clustered using an unsupervised clustering algorithm (hierarchical density-based spatial clustering of applications with noise) based on the spatial and numerical similarity of pixel values. Subsequent analysis revealed shifts in inter-residue distances and cluster size distributions across MTC treatment conditions. Notably, C25 induced the most significant structural changes, including localized disaggregation at the C- or N-terminal interactions, characterized by increased inter-residue distances and reduced cluster sizes. These structural alterations suggest that C25 disrupts tau dimer stabilization pathways that may underlie early aggregation. This clustering pipeline can also be applied to frame-by-frame data sets to detect rare protein events. Our findings may offer new insights into how MTCs disrupt early protein aggregation events in AD and other neurodegenerative diseases.