Cellulose dataset

Dataset: Multi-Scale Structural Embeddings of Crystalline and Amorphous Cellulose Description This dataset contains atomistic structures and calculated electronic properties of cellulose, specifically curated for training and validating machine learning models in computational materials science. The data supports the research presented at the MRS 2026 Spring Meeting regarding the development of transfer-learning architectures for molecular property prediction. The collection includes a diverse range of conformations, transitioning from highly ordered crystalline $(\text{I}\beta)$ structures to disordered amorphous states, obtained through a combination of Geometry Optimization and Molecular Dynamics (MD) simulations at 300 K. Dataset Specifications Total Configurations: [Insert Number, e.g., 5,000] unique frames. Chemical Scope: $\beta$-1,4-glucan chains in periodic and non-periodic boundary conditions. Ground Truth Data: * Total Energy and Atomic Forces (derived from [e.g., DFT/LAMMPS] calculations). Electronic Spectral Data: Full sets of Eigenvalues and Eigenvectors for structural embedding analysis. Format: Provided in [e.g., .xyz, .json, or .extxyz] for seamless integration with deep learning frameworks (PyTorch/TensorFlow). Methodology & Potential Use Cases The data was generated to facilitate the training of dual-head neural networks. It is particularly suited for: Transfer Learning: Fine-tuning foundational models (like ORB-v3) for organic polymer systems. Hybrid Loss Development: Testing optimization strategies involving MSE (for scalar properties) and Huber Loss (for vector/spectral properties). Phase Transition Studies: Analyzing the structural transition between crystalline and amorphous cellulose regions. How to Cite If you use this dataset in your research, please cite the associated MRS 2026 presentation: Rodríguez Aguirre, I., et al. "Predicting Electronic Properties of Complex Cellulose Systems via Fine-Tuned Structural Embeddings." MRS Spring Meeting 2026, Honolulu, HI.