google-deepmind/alphagenome_research — reverse-engineered prompt

Build me a Python research toolkit for AlphaGenome. I want it to let researchers load the AlphaGenome DNA model, download approved pretrained weights from Kaggle or Hugging Face after they accept the terms, and run predictions on long DNA intervals.

Make the main experience simple. A user should be able to create a model, enter a chromosome interval and a genetic variant, choose outputs like RNA expression, splicing, chromatin features, or contact maps, then get reference versus alternate predictions they can plot. Include helpers for variant scoring and in silico mutagenesis, plus a way to read training data from TFRecords.

Please include clear install instructions, a quick start example, and Colab notebooks that show a basic prediction and variant evaluation. Make it obvious that serious local runs need the right GPU setup, and suggest the hosted AlphaGenome API for people without that hardware. Look up current docs online if you need to.