Probing aggregation pathways of α-synuclein NAC_61–80 through machine learning-guided drug discovery and enhanced-sampling molecular dynamics

Parkinson’s disease arises from the aggregation of misfolded α-synuclein that impairs neurons. Oligomeric intermediates are highly toxic yet hard to probe because conventional molecular dynamics undersamples the conformational polymorphism of intrinsically disordered proteins. We combined generative machine learning with temperature replica-exchange molecular dynamics (REMD) to test how ligands perturb a 20-residue fragment from the aggregation-prone non-amyloid-β component (NAC_61–80). Focusing on early dimers, we trained a variational autoencoder on physicochemical objectives to generate a diverse ligand library, docked candidates to NAC_61–80, and simulated ligand-peptide assemblies with REMD. Ligand engagement induced clear structural shifts: bound systems occupied fewer, more stable states, and free-energy landscapes shifted away from compact, aggregation-prone conformations. Contact maps highlighted residues that seed oligomerization, with valine- and threonine-rich stretches as binding hotspots. Hydrogen bond analysis showed disruption of β-sheet–stabilizing interactions. These results clarify early NAC_61–80 assembly and identify targets to prevent aggregation.