Diffusion probabilistic modeling of protein backbones in 3D for the motif-scaffolding problem
Diffusion probabilistic modeling of protein backbones in 3D for the motif-scaffolding problem
复制标题
DOI：
10.48550/arxiv.2206.04119
复制
发表时间：
2022-06
期刊：
影响因子：
--
通讯作者：
Brian L. Trippe;Jason Yim;D. Tischer;Tamara Broderick;D. Baker;R. Barzilay;T. Jaakkola
中科院分区：
文献类型：
--
作者：
Brian L. Trippe;Jason Yim;D. Tischer;Tamara Broderick;D. Baker;R. Barzilay;T. Jaakkola
构建一个支架结构，支持所需的基序，赋予蛋白质功能，显示出疫苗和酶的设计的希望。但是这个主题支架问题的一般解决方案仍然是开放的。目前用于支架设计的机器学习技术要么局限于不切实际的小支架（最长20），要么难以生产多种不同的支架。我们建议通过E（3）-等变图神经网络来学习不同的和更长的蛋白质骨架结构的分布。我们开发了SMCDiff，以有效地从给定基序的条件下，从这个分布的支架样本;我们的算法是第一个从理论上保证条件样本从扩散模型在大计算限制。我们通过它们与AlphaFold 2预测结构的对齐程度来评估我们设计的主链。我们表明，我们的方法可以（1）样品支架多达80个残基和（2）实现结构多样的支架固定的基序。
Construction of a scaffold structure that supports a desired motif, conferring protein function, shows promise for the design of vaccines and enzymes. But a general solution to this motif-scaffolding problem remains open. Current machine-learning techniques for scaffold design are either limited to unrealistically small scaffolds (up to length 20) or struggle to produce multiple diverse scaffolds. We propose to learn a distribution over diverse and longer protein backbone structures via an E(3)-equivariant graph neural network. We develop SMCDiff to efficiently sample scaffolds from this distribution conditioned on a given motif; our algorithm is the first to theoretically guarantee conditional samples from a diffusion model in the large-compute limit. We evaluate our designed backbones by how well they align with AlphaFold2-predicted structures. We show that our method can (1) sample scaffolds up to 80 residues and (2) achieve structurally diverse scaffolds for a fixed motif.