NodeCoder: a graph-based machine learning platform to predict active sites of modeled protein structures
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:RMKCVA52record.jsonopen to challenge →
read the original abstract
While accurate protein structure predictions are now available for nearly every observed protein sequence, predicted structures lack much of the functional context offered by experimental structure determination. We address this gap with NodeCoder, a task-independent platform that maps residue-based datasets onto 3D protein structures, embeds the resulting structural feature into a contact network, and models residue classification tasks with a Graph Convolutional Network (GCN). We demonstrate the versatility of this strategy by modeling six separate tasks, with some labels derived from other experimental structure studies (ligand, peptide, ion, and nucleic acid binding sites) and other labels derived from annotation databases (post-translational modification and transmembrane regions). Moreover, A NodeCoder model trained to identify ligand binding site residues was able to outperform P2Rank, a widely-used software developed specifically for ligand binding site detection. NodeCoder is available as an open-source python package at https://pypi.org/project/NodeCoder/.
This paper has not been read by Pith yet.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.