5x Boost in In-Silico Drug Discovery with a Multimodal LLM

Working with the client’s in-silico drug discovery team, we developed a pipeline to train multi-modal LLMs with a deep understanding of chemistry. We integrated Llama 3.1 and Graphormer, incorporating molecular structures as graphs into the LLM, creating a unique model that could analyze textual, SMILES, and molecular data.