ESP

ESP predicts small-molecule substrates of enzymes using machine learning to identify enzyme–substrate pairs for enzymology and metabolic studies.


Key Features:

  • High Predictive Accuracy: Achieves over 91% accuracy on independent and diverse test datasets.
  • General Applicability: Predicts substrates across a wide range of enzymes rather than being limited to specific enzyme families.
  • Modified Transformer Enzyme Representation: Employs a modified transformer model to represent enzymes and capture complex patterns in enzyme–substrate interactions.
  • Data Augmentation via Negative Sampling: Addresses scarcity of negative examples by randomly sampling small molecules and assigning them as non-substrates to augment the training set.
  • Model Integration: Leverages both traditional machine learning and deep learning models for prediction.

Scientific Applications:

  • Basic Research: Generates hypotheses about enzyme substrate specificity and guides experimental investigation of enzymatic functions and metabolic pathways.
  • Drug Discovery and Metabolic Engineering: Suggests candidate substrates to aid identification of drug targets and to support optimization of engineered metabolic processes.

Methodology:

Training integrates traditional and deep learning models with a modified transformer-based enzyme representation on a dataset comprising known substrates and augmented non-substrate examples generated by random sampling of small molecules.

Topics

Details

Cost:
Free of charge
Tool Type:
web application
Operating Systems:
Mac, Linux, Windows
Added:
1/12/2024
Last Updated:
11/24/2024

Operations

Publications

Kroll A, Ranjan S, Engqvist MKM, Lercher MJ. A general model to predict small molecule substrates of enzymes based on machine and deep learning. Nature Communications. 2023;14(1). doi:10.1038/s41467-023-38347-2. PMID:37188731. PMCID:PMC10185530.

PMID: 37188731
Funding: - Deutsche Forschungsgemeinschaft: CRC 1310 + 390686111