9 December 2022 in Bioinformatics Advances
Machine learning-guided protein engineering is a rapidly advancing field. Despite major experimental and computational advances, collecting protein genotype (sequence) and phenotype (function) data remains time and resource intensive. As a result, the quality and quantity of training data is often a limiting factor in developing machine learning models. Data augmentation techniques have been successfully applied to the fields of computer vision and natural language processing, however, there is a lack of such augmentation techniques for biological sequence data. Towards this end we develop nucleotide augmentation (NTA), which leverages natural nucleotide codon degeneracy to augment protein sequence data via synonymous codon substitution. As a proof of concept for protein engineering, we test several online and offline augmentation implementations to train machine learning models with benchmark data sets of protein genotype and phenotype, revealing performance gains on par and surpassing benchmarks models using a fraction of the training data. NTA also enables substantial improvements for classification tasks under heavy class imbalance. The code used in this study is publicly available at https://github.com/minotm/NTA Supplementary data are available at Bioinformatics Advances online.
17 September 2025 in Cell Systems
The COVID-19 pandemic showcased a coevolutionary race between the human immune system and SARS-CoV-2, during which the immune system generated neutralizing antibodies targeting the SARS-CoV-2 spike protein's receptor-binding domain (RBD), crucial for host cell invasion, while the virus evolved to evade antibody recognition. Here, we establish a synthetic coevolution system combining high-throughput screening of antibody and RBD variant libraries with protein mutagenesis, surface display, and deep sequencing. Additionally, to significantly extend our interrogation of sequence space, we train a protein language model that predicts antibody escape to RBD variants and demonstrate its capability to generalize to a larger mutational load and mutations at positions unseen during training. Through explainable AI techniques, we probe the model and identify biologically meaningful coevolution trends. Synthetic coevolution reveals antagonistic and compensatory mutational trajectories of neutralizing antibodies and SARS-CoV-2 variants, enhancing the understanding of this evolutionary conflict.
30 January 2023 in bioRxiv
Machine learning-guided protein engineering continues to rapidly progress, however, collecting large, well-labeled data sets remains time and resource intensive. Directed evolution and protein engineering studies often require extensive experimental processes to eliminate noise and fully label high-throughput protein sequence-function data. Meta learning methods established in other fields (e.g. computer vision and natural language processing) have proven effective in learning from noisy data, given the availability of a small data set with trusted labels and thus could be applied for protein engineering. Here, we generate yeast display antibody mutagenesis libraries and screen them for target antigen binding followed by deep sequencing. Meta learning approaches are able to learn under high synthetic and experimental noise as well as in under labeled data settings, typically outperforming baselines significantly and often requiring a fraction of the training data. Thus, we demonstrate meta learning may expedite and improve machine learning-guided protein engineering. Availability and implementation The code used in this study is publicly available at https://github.com/LSSI-ETH/meta-learning-for-protein-engineering.