New Method Uses Interpretable AI Features to Predict Enzyme Functions in Microbial Proteins
Researchers developed a sparse autoencoder approach using the ESMC-6B protein language model to predict enzyme functions with 78.9% accuracy on known enzymes and identify novel enzyme candidates in microbial genomes. The method is interpretable by design, showing that predicted features correspond to known mechanistic concepts like catalytic triads and binding folds, and does not require GPU-intensive computation. This approach could accelerate discovery of enzymatic functions in the millions of uncharacterized microbial proteins, potentially unlocking new biotechnological applications.
Researchers have developed a new computational approach for predicting enzyme functions in microbial proteins by leveraging sparse autoencoder features from the ESMC-6B protein language model. The method achieved 78.9% top-1 accuracy on a benchmark of 4,868 microbial enzymes across 161 enzyme commission subclasses, substantially outperforming baseline sequence-similarity methods. Notably, the approach is interpretable—the learned features correspond to mechanistically meaningful biological concepts such as catalytic triad geometry in hydrolases and NAD(P)H-binding Rossmann folds in oxidoreductases. In a leave-one-EC3-class-out evaluation simulating discovery of novel enzyme classes, the method recovered the correct enzyme superclass in 47.7% of cases, 3.3 times better than random chance. The researchers also applied their method to 7.7 million protein clusters in the ESM Atlas and identified 169,859 candidate dark enzymes across microbial phyla. The approach is computationally efficient and does not require GPU-intensive training, making it scalable for large-scale protein discovery.
What's missing
The study does not discuss potential limitations of the sparse autoencoder approach, such as generalization to non-microbial proteins, performance on enzymes with unusual or rare catalytic mechanisms, or validation of the 169,859 predicted dark enzyme candidates through experimental methods. The reliance on GPT-5 annotations for feature interpretation and the reproducibility of those annotations is not addressed.
What different sources said
- arXiv q-bioCenter
Interpretable enzyme function prediction via sparse autoencoder features of ESMC across the microbial protein universe
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.