explore LLM for prediction

[kaichop]

we can evaluate how others have used BERT for the prediction (see the example code posted at kaggle), then evaluate how to create text representations possibly through RDkit to generate a rich representation of each chemical, and then assess performance of predictions.