Start with one of the largest protein language models, trained on hundreds of millions of proteins. Then finetune it on your own sequences, to make it specific to your classification task. Creates a GPU-backed API endpoint for classifying new sequences, retrieving probabilities, with your personalized ESM2 model. doi
Protein
Finetune a generative protein model on your own data, then create new sequences via API. Create de-novo best representative sequences that "follow the principles of natural ones and sample unexplored protein space". Only positive-class sequences required. doi
protein
DNA
DNA-based BERT models have demonstrated performance on promoter prediction, splice sites and transcription factor binding sites, up to 99% AUC. Bring your own NGS data or relevant public datasets to create any classifier. Your model also comes with an ability to export meaningful DNA BERT embeddings, perform semantic similarity searches. doi
dna