Accurate Protein Function Prediction via Graph Attention Networks with Predicted Structure Information

Abstract

Experimental protein function annotation does not scale with the fast-growing sequence databases. Only a tiny fraction (<0.1%) of protein sequences in UniProtKB has experimentally determined functional annotations. Computational methods may predict protein function in a high-throughput way, but its accuracy is not very satisfactory. Based upon recent breakthroughs in protein structure prediction and protein language models, we develop GAT-GO, a graph attention network (GAT) method that may substantially improve protein function prediction by leveraging predicted inter-residue contact graphs and protein sequence embedding. Our experimental results show that GAT-GO greatly outperforms the latest sequence- and structure-based deep learning methods. On the PDB-mmseqs testset where the train and test proteins share <15% sequence identity, GAT-GO yields Fmax(maximum F-score) 0.508, 0.416, 0.501, and AUPRC(area under the precision-recall curve) 0.427, 0.253, 0.411 for the MFO, BPO, CCO ontology domains, respectively, much better than homology-based method BLAST (Fmax 0.117,0.121,0.207 and AUPRC 0.120, 0.120, 0.163). On the PDB-cdhit testset where the training and test proteins share higher sequence identity, GAT-GO obtains Fmax 0.637, 0.501, 0.542 for the MFO, BPO, CCO ontology domains, respectively, and AUPRC 0.662, 0.384, 0.481, significantly exceeding the just-published graph convolution method DeepFRI, which has Fmax 0.542, 0.425, 0.424 and AUPRC 0.313, 0.159, 0.193.

NetGO 3.0: Protein Language Model Improves Large-scale Functional Annotations

“…We have developed NetGO 3.0 to improve the performance of large-scale AFP by incorporating a new component LR-ESM, which utilizes a protein language model to generate powerful representations of proteins. Interesting future work would be integrating protein structural information into NetGO to enhance the performance of AFP [18][19][20]. Each GO term is attached with tags, which illustrates that the GO term is predicted correctly by corresponding methods.…”

BayeStab: Predicting effects of mutations on protein stability with uncertainty quantification

“…Graph convolutional networks apply spectral convolution in the graph Fourier domain to aggregate neighboring representations for feature learning 32 . They have been used for protein structure refinement 33 and protein function prediction 34 . These attempts to encode protein context information make the prediction of mutation induced stability changes possible, yet it is still scarcely investigated.…”

“…32 They have been used for protein structure refinement 33 and protein function prediction. 34 These attempts to encode protein context information make the prediction of mutation induced stability changes possible, yet it is still scarcely investigated.…”

A Deep SE(3)-Equivariant Model for Learning Inverse Protein Folding

“…As protein function is jointly determined by its 3-dimensional confor-mation and primary amino acid sequence (Mitchell et al 2019; Dawson et al 2017) this suggests that functional effects of mutations should be modeled jointly as well. It has been shown that protein structure information can be used to predict its function (Gligorijevic et al 2021; Lai and Xu 2022). Although the most successful existing methods for protein mutation effect prediction rely on sequence and evolutionary information (Meier et al 2021; Hopf et al 2017; Riesselman et al 2018), we hypothesized that a generative model, conditioned solely on structure and partial sequence, could be used as a zero-shot predictor for functional effects on single point mutations.…”