Evo: DNA foundation modeling from molecular to genome scale

Here is a colab notebook that replicates some of their examples without needing a togetherAI API: https://colab.research.google.com/drive/1pzz1KHQjylrpfZt2xS7...

Choosing the A100 environment is required though

I've often thought that an interesting application of single-nucleotide resolution, large context-length models would be to train them on GWAS data to predict phenotypes from large DNA sequences (possibly the whole genome) and see if the model can perform better than current polygenetic scores. I'm not a geneticist nor an expert on machine learning, so I don't know how feasible this is given current technology and datasets.

Trained on 80,000 propagate sequences.

There's a long way to go for these types of approaches