Protein model quality assessment using rotation-equivariant,\n hierarchical neural networks

2020/11/27 by Stephan Eismann, Eismann, Stephan, Patricia Suriana +7
Biochemistry, Genetics and Molecular Biology · Materials Science · #Enzyme Structure and Function #FOS: Biological sciences #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning in Bioinformatics #Machine Learning in Materials Science #Protein Structure and Dynamics #Quantitative Methods (q-bio.QM)

paper · pdf · doi:10.48550/arxiv.2011.13557

openalex publication_date 2020/11/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Proteins are miniature machines whose function depends on their\nthree-dimensional (3D) structure. Determining this structure computationally\nremains an unsolved grand challenge. A major bottleneck involves selecting the\nmost accurate structural model among a large pool of candidates, a task\naddressed in model quality assessment. Here, we present a novel deep learning\napproach to assess the quality of a protein model. Our network builds on a\npoint-based representation of the atomic structure and rotation-equivariant\nconvolutions at different levels of structural resolution. These combined\naspects allow the network to learn end-to-end from entire protein structures.\nOur method achieves state-of-the-art results in scoring protein models\nsubmitted to recent rounds of CASP, a blind prediction community experiment.\nParticularly striking is that our method does not use physics-inspired energy\nterms and does not rely on the availability of additional information (beyond\nthe atomic structure of the individual protein model), such as sequence\nalignments of multiple proteins.\n

Citations

• Dictionary of protein secondary structure: Pattern recognition of hydrogen‐bonded and geometrical features
• Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
• Improved protein structure prediction using potentials from deep learning
• Hierarchical, rotation‐equivariant neural networks to select structural models of protein complexes

Related

• Deep transfer learning in the assessment of the quality of protein models
• DeepQA: Improving the estimation of single protein model quality with deep belief networks
• Hierarchical, rotation‐equivariant neural networks to select structural models of protein complexes
• Spherical convolutions on molecular graphs for protein model quality assessment
• ProQ3D: Improved model quality assessments using Deep Learning