Deep learning methods enable the structural prediction of proteins with high accuracy but are unable to model non-protein molecules that are essential for a protein’s biological function. Writing in Science, Krishna et al. introduce RoseTTAFold All-Atom (RFAA) to model the structure of full biological assemblies containing proteins, nucleic acids, small molecules, metals and covalent modifications.

A challenge in modeling generalized biomolecular assemblies is how to present all the components. Whereas proteins and nucleic acids can be modeled as linear chains, many small molecules are not polymers and need different representations. The authors tackle this problem by building RFAA using sequence-based representations of biopolymers combined with an atomic graph representation of small molecules and covalent modifications.

深度学习方法可以高精度地预测蛋白质的结构，但无法对蛋白质生物功能所必需的非蛋白质分子进行建模。克里希纳等人在《科学》上写作。引入 RoseTTAFold All-Atom (RFAA) 来模拟包含蛋白质、核酸、小分子、金属和共价修饰的完整生物组装体的结构。

广义生物分子组装建模的一个挑战是如何呈现所有组件。虽然蛋白质和核酸可以建模为线性链，但许多小分子不是聚合物，需要不同的表示。作者通过使用基于序列的生物聚合物表示结合小分子和共价修饰的原子图表示构建 RFAA 来解决这个问题。