Predicting the tertiary structure of a protein from its primary sequence has been greatly improved by integrating deep learning and co-evolutionary analysis, as shown in CASP13 and CASP14. We describe our latest study of this idea, analysing the efficacy of network size and co-evolution data and its performance on both natural and designed proteins. We show that a large ResNet (convolutional residual neural networks) can predict structures of correct folds for 26 out of 32 CASP13 free-modelling targets and L/5 long-range contacts with precision over 80%. When co-evolution is not used, ResNet can still predict structures of correct folds for 18 CASP13 free-modelling targets, greatly exceeding previous methods that do not use co-evolution either. Even with only the primary sequence, ResNet can predict the structures of correct folds for all tested human-designed proteins. In addition, ResNet may fare better for the designed proteins when trained without co-evolution than with co-evolution. These results suggest that ResNet does not simply de-noise co-evolution signals, but instead may learn important protein sequence–structure relationships. This has important implications for protein design and engineering, especially when co-evolutionary data are unavailable.

通过整合深度学习和协同进化分析，从蛋白质的一级序列预测蛋白质的三级结构得到了极大的改进，如 CASP13 和 CASP14 所示。我们描述了我们对这一想法的最新研究，分析了网络大小和共同进化数据的功效及其对天然和设计蛋白质的性能。我们表明，大型 ResNet（卷积残差神经网络）可以预测 32 个 CASP13 自由建模目标中的 26 个和 L/5 远程接触的正确折叠结构，精度超过 80%。在不使用协同进化的情况下，ResNet 仍然可以预测 18 个 CASP13 自由建模目标的正确折叠结构，大大超过了以前也不使用协同进化的方法。即使只有一级序列，ResNet 可以预测所有经过测试的人类设计蛋白质的正确折叠结构。此外，在没有协同进化的情况下训练时，ResNet 对于设计的蛋白质可能比协同进化的情况更好。这些结果表明，ResNet 不仅可以对协同进化信号进行去噪，还可以学习重要的蛋白质序列-结构关系。这对蛋白质设计和工程具有重要意义，尤其是在无法获得共同进化数据的情况下。