A language model assigns probabilities to sequences or predicts the next token from preceding tokens. In drug discovery, a related kind of sequence-trained foundation model can produce representations called embeddings for downstream tasks. section 13.1.2
Protein-sequence embeddings
A foundation model trained on protein sequences can generate embeddings used for structure prediction and function annotation. Information about a binding-site environment can also help guide molecular generation. A sequence-trained foundation model is not necessarily a language model. section 13.1.2
























































































