AutoGraph

AutoGraph performs autonomous graph-based clustering of metabolite conformers to reduce redundancy in configuration space and organize conformational ensembles for modeling molecular flexibility.

Key Features:

• Autonomous graph-based clustering: Performs graph-based clustering of metabolite conformers to reduce redundancy in configuration space.
• Louvain algorithm: Implements the Louvain algorithm for conformational clustering.
• Minimal hyperparameters: Minimizes reliance on user-defined hyperparameters such as the number of clusters or threshold values.
• Preservation of intrinsic relationships: Preserves intrinsic geometric and energetic relationships within conformational data without imposing arbitrary constraints.
• PES-aligned partitioning: Partitions ensembles based on geometric and energetic correlations that align with points on the potential energy surface.
• Conformational ensemble generation: Includes generation, refinement, and clustering of conformations to produce comprehensive conformational ensembles.
• Metabolite validation: Applied to a set of 200 representative metabolites including O-succinyl-L-homoserine and oxidized nicotinamide adenine dinucleotide.

Scientific Applications:

• Modeling molecular flexibility: Characterizes conformational ensembles for flexible molecules whose properties depend on conformational states.
• Redundancy reduction: Reduces redundancy in conformational configuration space for downstream computational or experimental analysis.
• Reproducible clustering: Mitigates human bias from manual hyperparameter selection to improve reproducibility of conformer clustering.
• Large-scale metabolite analysis: Enables conformational analysis across metabolite libraries, demonstrated on 200 representative metabolites.
• Mapping to potential energy surface: Relates conformational clusters to points on the potential energy surface via geometric and energetic correlations.

Methodology:

Generating, refining, and clustering conformations using graph-based representations and the Louvain algorithm, with partitioning informed by geometric and energetic correlations aligned to the potential energy surface.