Crystal Structure Determination from Powder Diffraction Patterns with Generative Machine Learning.

Powder X-ray diffraction (PXRD) is a cornerstone technique in materials characterization. However, complete structure determination from PXRD patterns alone remains time-consuming and is often intractable, especially for novel materials. Current machine learning (ML) approaches to PXRD analysis predict only a subset of the total information that comprises a crystal structure. We developed a pioneering generative ML model designed to solve crystal structures from real-world experimental PXRD data. In addition to strong performance on simulated diffraction patterns, we demonstrate full structure solutions over a large set of experimental diffraction patterns. Benchmarking our model, we predicted the structure for 134 experimental patterns from the RRUFF database and thousands of simulated patterns from the Materials Project on which our model achieves state-of-the-art 42 and 67% match rate, respectively. Further, we applied our model to determine the unreported structures of materials such as NaCu2P2, Ca2MnTeO6, ZrGe6Ni6, LuOF, and HoNdV2O8 from the Powder Diffraction File database. We extended this methodology to new materials created in our lab at high pressure with previously unsolved structures and found the new binary compounds Rh3Bi, RuBi2, and KBi3. We expect that our model will open avenues toward materials discovery under conditions which preclude single crystal growth and toward automated materials discovery pipelines, opening the door to new domains of chemistry.

An all-atom protein generative model.

Proteins mediate their functions through chemical interactions; modeling these interactions, which are typically through sidechains, is an important need in protein design. However, constructing an all-atom generative model requires an appropriate scheme for managing the jointly continuous and discrete nature of proteins encoded in the structure and sequence. We describe an all-atom diffusion model of protein structure, Protpardelle, which represents all sidechain states at once as a "superposition" state; superpositions defining a protein are collapsed into individual residue types and conformations during sample generation. When combined with sequence design methods, our model is able to codesign all-atom protein structure and sequence. Generated proteins are of good quality under the typical quality, diversity, and novelty metrics, and sidechains reproduce the chemical features and behavior of natural proteins. Finally, we explore the potential of our model to conduct all-atom protein design and scaffold functional motifs in a backbone- and rotamer-free way.

An all-atom protein generative model.

Proteins mediate their functions through chemical interactions; modeling these interactions, which are typically through sidechains, is an important need in protein design. However, constructing an all-atom generative model requires an appropriate scheme for managing the jointly continuous and discrete nature of proteins encoded in the structure and sequence. We describe an all-atom diffusion model of protein structure, Protpardelle, which instantiates a "superposition" over the possible sidechain states, and collapses it to conduct reverse diffusion for sample generation. When combined with sequence design methods, our model is able to co-design all-atom protein structure and sequence. Generated proteins are of good quality under the typical quality, diversity, and novelty metrics, and sidechains reproduce the chemical features and behavior of natural proteins. Finally, we explore the potential of our model conduct all-atom protein design and scaffold functional motifs in a backbone- and rotamer-free way.