Researchers from the University of Washington Developed a Deep Learning Method for Protein Sequence Design that Explicitly Models the Full Non-Protein Atomic Context