Protein Sequence and Structure Co-Design with Equivariant Translation

Shi, Chence; Wang, Chuanrui; Lu, Jiarui; Zhong, Bozitao; Tang, J.

doi:10.48550/arxiv.2210.08761

Cited by 5 publications

(6 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…C α C β 而在ProteinMPNN [108] 中, 研究者们则使用了图神经网络(graph neural networks, GNN) [109] 的框架, 如图 4所示. 常是在设计分子间相互作用下的蛋白质 [110,111] 时所面对的.…”

Section: 结构空间中通过寻找能量最低点的方法找到预测unclassified

Machine learning for <i>in silico</i> protein research

Zhang

2024

Acta Phys. Sin.

View full text Add to dashboard Cite

In silico protein research has been a focus for a long time, while its recent combination with machine learning gives great contributions to related areas. This review mainly focuses on four major fields of in silico protein research that combine with machine learning, which are molecular dynamics, structure prediction, property prediction and molecule design. Molecular dynamics depends on force field parameters, which is necessary for accurate results. Machine learning can help to construct satisfied force field parameters. In molecular dynamics, machine learning can also help to do the free energy calculation by a relatively low cost. Structure prediction is usually predicting the structure of given sequence. Structure prediction is of high complexity and data amount, which is what machine learning good at. By the help of machine learning, scientists have gained great achievements in 3D structure prediction of proteins. On the other hand, predicting proteins’ properties based on their known information is also important to protein research. The most challenging, however, is molecule design. Though drug-like small molecule design and protein design have achieved much in recent years, this field is still remain to be explored. This review focuses on the above three areas and make an outlook to in silico protein research with machine learning.

show abstract

Section: 结构空间中通过寻找能量最低点的方法找到预测unclassified

Machine learning for <i>in silico</i> protein research

Zhang

2024

Acta Phys. Sin.

View full text Add to dashboard Cite

show abstract

“…McPartlon et al (2022) replaces GVP with an SE(3)-equivariant transformer. Dauparas et al (2022) and Shi et al (2022) further improve performance on the inverse folding task by further optimizing the architecture and decoding schemes. MIF-ST uses sequence transfer via a sequence masked language model to improve MIF.…”

Section: Related Workmentioning

confidence: 99%

“…Meanwhile, inverse folding methods reconstruct a protein's amino-acid sequence given its structure. Deep learning-based inverse folding is usually parametrized as a graph neural network (GNN) (Ingraham et al, 2019;Strokach et al, 2020;Jin et al, 2021;Jing et al, 2020;Hsu et al, 2022;Dauparas et al, 2022;Shi et al, 2022) or SE(3)-equivariant transformer (McPartlon et al, 2022) that either reconstructs or autoregressively decodes the amino-acid sequence conditioned on the desired backbone structure. This is similar in spirit to work in fixed-backbone protein design, which involves the design of proteins with a given target backbone structure.…”

Section: Introductionmentioning

confidence: 99%

Masked Inverse Folding with Sequence Transfer for Protein Representation Learning

Yang

Zanichelli

Yeh

2022

Preprint

View full text Add to dashboard Cite

Self-supervised pretraining on protein sequences has led to state-of-the art performance on protein function and fitness prediction. However, sequence-only methods ignore the rich information contained in experimental and predicted protein structures. Meanwhile, inverse folding methods reconstruct a protein’s amino-acid sequence given its structure, but do not take advantage of sequences that do not have known structures. In this study, we train a masked inverse folding protein language model parameterized as a structured graph neural network. We then show that using the outputs from a pretrained sequence-only protein masked language model as input to the inverse folding model further improves pretraining perplexity. We evaluate both of these models on downstream protein engineering tasks and analyze the effect of using information from experimental or predicted structures on performance.

show abstract

“…There are many protein and antibody design programs [11][12] , such as diffAb 13 , PROTSEED 14 , Rosetta [15][16][17] , RFdiffusion [18][19] , SCUBA 20 , Abacus-R 21 , Fold-X 22 , EGAD [23][24] , 3DCNN [25][26] etc. All of these have the potential to design antibodies targeting specific epitopes.…”

Section: Introductionmentioning

confidence: 99%

Combining transformer and 3DCNN models to achieve co-design of structures and sequences of antibodies in a diffusional manner

Hu,

Tao,

Lan

et al. 2024

Preprint

View full text Add to dashboard Cite

Antibody drugs are among the fastest growing therapeutic modalities in modern drug research and development. Due to the huge search space of antibody sequences, the traditional experimental screening method cannot fully meet the needs of antibody discover. More and more rational design methods have been proposed to improve the success rate of antibody drugs. In recent years, artificial intelligence methods have increasingly become an important means of rational design. We have proposed an algorithm for antibody design, called AlphaPanda (AlphaFold2 inspired Protein-specific antibody design in a diffusional manner). The algorithm mainly combines the transformer model, the 3DCNN model and the diffusion generative model, use the transformer model to capture the global information and uses the 3DCNN model to capture the local structural characteristics of the antibody-antigen complexes, and then uses the diffusion model to generate sequences and structures of antibodies. The 3DCNN model can capture pairwise interactions in antibody-antigen complex, as well as non-pairwise interactions in antibody-antigen complex, and it requires less training sample data, while avoiding the defects of the generation progress by the autoregressive model and by the self-consistent iterative model. Diffusion generative model can generate sequence and structure effectively and with high quality. By combining 3DCNN method and diffusion model method, we have achieved the integration of 3DCNN model to the protein design with flexible main chains. By utilizing the advantages of these aspects, a good performance has been achieved by the AlphaPanda algorithm. The algorithm we propose can not only be applied to antibody design, but also be more widely applied to various fields of other protein design. The source code can be get from github (https://github.com/YueHuLab/AlphaPanda).

show abstract

Protein Sequence and Structure Co-Design with Equivariant Translation

Cited by 5 publications

References 25 publications

Machine learning for <i>in silico</i> protein research

Machine learning for <i>in silico</i> protein research

Masked Inverse Folding with Sequence Transfer for Protein Representation Learning

Combining transformer and 3DCNN models to achieve co-design of structures and sequences of antibodies in a diffusional manner

Contact Info

Product

Resources

About