Authors:

Joel Jenson Kuriakose

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

This study investigated data-driven approaches for predicting the function of enzymes using primary amino acid sequences. Our long-term goal is to predict the substrates that bind to enzymes and the reactions catalyzed by enzymes using primary amino acid sequences alone. Here, we classify enzymes based on their ability to bind NADPH/NADH family of cofactors, and it represents an important first step towards our long-term goal. The study began with the curation of a dataset of approximately 360,000 sequences with corresponding reaction descriptions from RHEA/UniProt databases. This dataset comprises diverse enzymatic transformations seen in Nature. Relevant reactions that use NADH/NADPH were then identified by filtering the reactions to identify ones containing these molecules. Data cleaning steps were crucial; this included eliminating duplicates, disregarding sequences containing non-standard amino acids, and removing low- frequency occurrences. Our final dataset comprises 31,329 positive sequences that bind to NADH/NADPH family of cofactors and 31,329 negative sequences that do not bind to these cofactors. A transformer model was then trained to predict the capability of an input amino acid sequence to bind to the NADH/NADPH family of cofactors. On a hidden test set comprising of 6,266 amino acid sequences, the model achieved 82% accuracy on discriminating between sequences that bound to NADPH/NADH and sequences that do not bind these cofactors. This study opens the door for computational substrate and reaction prediction using primary amino acid sequences. Such tools can facilitate identification of biocatalysts to catalyze desired industrial transformations. Keywords: [no keywords provided]

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Joel Jenson Kuriakose