Nucleotide augmentation for machine learning-guided protein engineering.

Area of Science:

• Biotechnology
• Computational Biology
• Machine Learning

Background:

• Protein engineering relies on machine learning but is limited by the time and resource-intensive collection of protein sequence and function data.
• Existing data augmentation techniques are not readily applicable to biological sequence data.

Purpose of the Study:

• To develop a novel data augmentation technique for protein sequence data.
• To address the limitations of data scarcity in machine learning for protein engineering.

Main Methods:

• Developed Nucleotide Augmentation (NTA) leveraging synonymous codon substitution.
• Implemented and tested online and offline NTA methods for augmenting protein genotype-phenotype datasets.
• Evaluated NTA performance on benchmark datasets for protein engineering tasks.

Main Results:

• NTA demonstrated performance gains comparable to or exceeding benchmark models trained on larger datasets.
• NTA achieved these results using a significantly smaller fraction of the training data.
• NTA significantly improved classification performance, especially in scenarios with imbalanced classes.

Conclusions:

• Nucleotide Augmentation (NTA) is an effective method for increasing the quantity and quality of training data in machine learning for protein engineering.
• NTA offers a viable solution to data scarcity challenges, enabling more robust and accurate protein engineering models.
• The developed NTA approach shows promise for various machine learning applications involving biological sequence data.