summary: Researchers have used artificial intelligence to predict how enzymes will interact with different substrates. The research team has developed an AI model that can accurately predict whether an enzyme can work with a particular molecule.
Their Enzyme Substrate Prediction (ESP) model provides a valuable tool for pharmaceutical research and biotechnology, with applications ranging from drug discovery to biofuel production.
Important facts:
- The developed AI-based method can predict whether an enzyme can interact with a particular molecule with 91% accuracy.
- The ESP model works with any combination of enzymes and over 1,000 different substrates.
- The methods developed are useful in pharmaceutical research, biotechnology, cellular metabolism simulations, and help to understand the physiology of various organisms.
sauce: Heinrich Heine University Düsseldorf
Enzymes are molecular factories within living cells. However, what basic molecular building blocks are used to assemble the target molecule is often unknown and difficult to measure.
An international team including bioinformaticians from the Heinrich Heine University (HHU) in Düsseldorf has taken an important step in this regard. Their AI method predicts with high accuracy whether an enzyme will work on a particular substrate.
They are now publishing their results in a scientific journal Nature Communications.
Enzymes are important biocatalysts in all living cells. Enzymes facilitate chemical reactions by which all molecules important to living organisms are produced from basic substances (substrates). Most organisms have thousands of different enzymes, each responsible for a very specific reaction. The collective function of all enzymes constitutes metabolism and provides the conditions for life and survival of organisms.
Genes encoding enzymes are readily identifiable per se, but the exact function of the resulting enzymes is in most cases (more than 99%) unknown. This is because experimental characterization of their function (i.e. which starting molecule a particular enzyme converts to which concrete final molecule) is very time consuming.
A research team led by Professor Martin Lercher of the HHU Computational Cell Biology Research Group collaborated with colleagues in Sweden and India to develop an AI-based method to predict whether an enzyme can use a given molecule as a substrate for a reaction. Did. catalyze
Prof. Lurcher: “A feature of our ESP (‘Enzyme Substrate Prediction’) model is that it is not limited to individual specialized enzymes and other closely related enzymes, as was the case in previous models. Our general model works with any combination of enzymes and over 1,000 different substrates. ”
PhD student Alexander Kroll, lead author of the study, developed a so-called deep learning model in which information about enzymes and substrates is encoded in mathematical structures known as numeric vectors.
A vector of approximately 18,000 experimentally validated enzyme-substrate pairs (enzymes and substrates are known to work together) were used as inputs to train a deep learning model.
Alexander Kroll: “After training the model in this way, we applied it to an independent test dataset for which we already knew the correct answer. was accurately predicted.”
This method offers a wide range of potential applications. Knowing which substances are transformed by enzymes is very important in both drug discovery research and biotechnology.
Prof. Lurcher: “This will allow research and industry to narrow down a large number of possible combinations to the most promising ones, which can then be used for enzymatic production of new drugs, chemicals and even biofuels. .”
Kroll adds: “It will also enable the creation of improved models that simulate cellular metabolism. It will also help us understand the physiology of organisms ranging from bacteria to humans.”
In addition to Kroll and Lercher, Professor Martin Enqvist of Chalmers University of Technology, Gothenburg, Sweden, and Sahasra Ranjan of the Indian Institute of Technology, Mumbai, also participated in the study. Engqvist helped design the study, and Ranjan implemented a model that encodes enzyme information that feeds into the overall model developed by his Kroll.
About this Artificial Intelligence Research News
author: Arne Clausen
sauce: Heinrich Heine University Düsseldorf
contact: Arne Clausen – Heinrich Heine University Düsseldorf
image: Image credited to Neuroscience News
Original research: open access.
“Enzyme Substrate Range: A General Predictive Model Based on Machine Learning and Deep Learning” Martin Lercher et al. Nature Communications
overview
Enzyme Substrate Range: A Common Predictive Model Based on Machine Learning and Deep Learning
For most proteins annotated as enzymes, it is unknown which primary and/or secondary reactions they catalyze. Experimental characterization of potential substrates is time consuming and expensive. Machine learning prediction may provide an efficient alternative, but it is hampered by the lack of information on non-enzymatic substrates as the available training data consist mainly of positive examples. .
Here, we present ESP, a popular machine learning model for predicting enzyme-substrate pairs with >91% accuracy on independent and diverse test data.
ESP can be applied to a wide variety of enzymes and a wide range of metabolites in the training data and outperforms models designed for well-studied individual enzyme families.
ESP represents the enzyme through a modified transformer model and is trained on data enriched with randomly sampled small molecules assigned as non-substrates.
By facilitating in silico testing of potential substrates, ESP web servers can support both basic and applied sciences.
