Gene Ontology annotation based on Enzyme Commission mapping
Annotation inferences using phylogenetic trees
Automatic Gene Ontology annotation based on Rhea mapping
Electronic Gene Ontology annotations created by ARBA machine learning models
Interaction network containing conserved and essential protein complexes in Escherichia coli.
-
High-throughput affinity purification-mass spectrometry detected a physical interaction between YgfF (P52037) and LpdA (P0A9P0, dihydrolipoyl dehydrogenase).
"no large-scale analysis of protein complexes in Escherichia coli has yet been reported. To this end, we have targeted DNA cassettes into the E. coli chromosome to create carboxy-terminal, affinity-tagged alleles of 1,000 open reading frames"
Functional annotation of enzyme-encoding genes using deep learning with transformer layers.
-
DeepECtransformer predicted YgfF to have EC number EC:1.1.1.47 (glucose 1-dehydrogenase) and this was experimentally validated by in vitro enzyme assay with a specific activity of 305.55 U/mg.
"For YgfF, DeepECtransformer predicted its EC number to be EC:1.1.1.47 (glucose 1-dehydrogenase). The enzyme assay results showed that YgfF exhibited a specific glucose 1-dehydrogenase activity of 305.55 U mg−1"
-
YgfF was one of three randomly selected y-ome proteins whose predicted enzymatic functions were experimentally validated, demonstrating the utility of DeepECtransformer for functional annotation.
"we randomly selected three proteins, YgfF, YciO, and YdjM, that are predicted to be oxidoreductase, transferase, and hydrolase, respectively"
-
The neural network predicted YgfF's function despite higher sequence identity to a different enzyme (EC:1.1.1.100), showing the model learned functional motifs rather than relying solely on homology.
"It should be noted that although YgfF exhibited a higher sequence identity with a different enzyme (A0A069CGU9_ECOLX; EC:1.1.1.100) from the training dataset than glucose 1-dehydrogenase exhibiting the maximum sequence identity within the training dataset, the neural network made an accurate prediction"
Back to the future of metabolism - advances in the discovery and characterization of unknown biocatalytic functions and pathways.
-
YgfF is highlighted as an example of ML-assisted functional annotation where DeepECtransformer predicted glucose 1-dehydrogenase activity and in vitro enzyme assays were performed on overexpressed and affinity-purified protein. The review emphasizes that deeper characterization (kinetic parameters, substrate spectrum) is needed to fully establish metabolic roles.
"The review stresses that function assignment for unusual/unknown enzymes often requires extensive experimental work including expression, purification, substrate synthesis, analytical methods, and kinetic characterization such as kcat and KM."
Limitations of current machine-learning models in predicting enzymatic functions for uncharacterized proteins.
-
YgfF is classified into the SDR63C / glucose 1-dehydrogenase subgroup by HMM-based SDR subfamily analysis, consistent with the DeepECtransformer prediction. However, in vitro activity alone is insufficient to establish physiological (in vivo) function, and best practice requires combining biochemical and genetic evidence.
"This resource predicts YgfF is part of the SDR63C/Glucose 1-dehydrogenase subgroup, the activity predicted and validated in the Kim et al. (2023) study. This prediction demonstrates the accurate propagation of functional annotation and is a successful prediction."