Deep Learning for Molecules and Materials (dmol.pub)

"Deep Learning for Molecules and Materials" is a pedagogical, code-first textbook (White 2021, Living Journal of Computational Molecular Science) that frames deep learning as a practical, now-standard tool for chemistry and materials science. The book argues deep nets enable previously infeasible models—accelerating quantum calculations to near-DFT accuracy and supporting generative design—by removing hand-crafted descriptors through end-to-end learning. It emphasizes mature ecosystems (automatic differentiation, GPUs) that make experimenting and scaling models tractable, while stressing that deep learning complements—not replaces—domain expertise. Targeted at students with chemistry and Python backgrounds, the book pairs math, algorithmic intuition, and runnable examples (interactive Google Colabs). It focuses on architectures and tools most useful to the field: Graph Neural Networks and Variational Autoencoders for structure and generation; concrete implementations such as SchNet for space-group prediction, graph convolutional nets for DFT single-point energies, and RNN-based molecule generation; plus simpler baselines (logistic regression for toxicity). Implementation uses Jax for low-level clarity, Keras/TensorFlow (and TensorFlow Probability) for higher-level models, and scikit-learn for classic ML; concepts are presented framework-agnostically so readers can transfer to PyTorch or others. The book includes practical GPU guidance, community-contributed code/math fixes, and is published under CC BY-NC with NIH/NSF-supported work.