Intrinsically disordered proteins (IDPs) exhibit highly dynamic and heterogeneous conformational ensembles that are strongly influenced by sequence features. While global properties such as chain compaction and scaling behavior have been widely studied, they often fail to resolve the fine-grained, sequence-specific structural variation that underlies IDP function. Here, we perform long-time scale atomistic simulations of 47 representative IDP sequences from the yeast proteome to systematically investigate the relationship between sequence composition and conformational ensemble. To analyze the high-dimensional structural data, we apply uniform manifold approximation and projection (UMAP), a nonlinear dimensionality reduction technique that preserves local structural relationships. The resulting low-dimensional embeddings effectively differentiate IDP ensembles and reveal a novel descriptor─local compactness asymmetry─that quantifies directional differences in chain organization. This metric, denoted ${\gamma }_{R_g}$, captures conformational features orthogonal to traditional global measures such as radius of gyration and end-to-end distance. We show that ${\gamma }_{R_g}$ correlates with sequence-level asymmetries in charge and hydropathy, and that conformational dynamics preferentially occur in the more extended region of the chain. The simulation data set generated in this work also provides a valuable resource for training machine learning models and developing improved coarse-grained force fields for disordered proteins.