LLM for recommending cancer modeling publications

[lorikrammer]

We would like to have a PredictMod LLM that recommends PubMed articles for cancer ML modeling.

Once Miguel has completed #328, Sean or Robel will use the abstracts identified to train/prompt an LLM which will make recommendations from the PubMed query of 20,000 abstracts.

This effort should take <8 hours to train the model, and 16 hours for a human to evaluate the model performance.

[lorikrammer]

Miguel has provided the list of pubmed articles and query parameters in #328.

[lorikrammer]

@seankim658 to prototype this and train @mariacuria

[seankim658]

@lorikrammer Correct me if I'm wrong, but based on the Friday conversation my understanding is that we are just using OpenAI apis to read abstracts and tell us if it is potentially a good source for ML-ready cancer data?

[lorikrammer]

@seankim658 Yes that's correct. You'll prompt the LLM to consider articles that we have established as good for modeling based off of the articles Miguel recommended.

[lorikrammer]

Hi @mariacuria I know you wanted to touch base on this item before I was out of office. I'm happy to meet to discuss!