PETra
Pretrained Evolutionary Transformer for SARS-CoV-2 mutation prediction. Published on arXiv 2025 (sole author).
PETra (Pretrained Evolutionary Transformer) is a deep learning model for predicting future mutations in SARS-CoV-2. It leverages a transformer architecture pretrained on large-scale viral genomic data to capture evolutionary patterns and forecast which amino acid substitutions are most likely to emerge.
Unlike traditional phylogenetic methods that rely on tree-based inference, PETra frames mutation prediction as a sequence modeling problem: given the evolutionary history of the virus, what mutations are likely to appear next?
This work was published as a sole-author paper on arXiv (2025):
“PETRA: Pretrained Evolutionary Transformer for SARS-CoV-2 Mutation Prediction” (arXiv:2511.03976).
Visit PETra for more details.