Google Research and Google DeepMind have introduced a groundbreaking large language model (LLM) for drug discovery and therapeutic development, named Tx-LLM, which is fine-tuned from the company’s PaLM-2 model.
Tx-LLM leverages PaLM-2, Google’s advanced generative AI technology, which has been designed to address complex medical queries. This specialized LLM has been trained using 709 datasets, covering 66 critical tasks throughout the drug discovery process. These tasks include assessing efficacy and safety, predicting therapeutic targets, and evaluating manufacturing feasibility.
How Tx-LLM Works
The model integrates the Therapeutics instruction Tuning (TxT) collection, a dataset that combines free-text instructions with representations of small molecules, such as SMILES strings (Simplified Molecular Input Line Entry System). SMILES is a notation system that uses printable characters to represent chemical structures and reactions.
TxT serves as a key component in fine-tuning Tx-LLM, enabling it to excel in classification, regression, and generative tasks related to drug discovery. For instance, when predicting drug synergy, researchers crafted prompts that included instructions, relevant context, and specific questions to guide the model.
Performance and Findings
Tx-LLM demonstrated exceptional performance, surpassing or matching state-of-the-art (SOTA) models in 43 of 66 tasks and outperforming SOTA models in 22 tasks. The researchers also observed positive data transfer effects, where training on datasets with diverse drug types, including biological sequences, enhanced performance on molecular datasets.
“The proposed Tx-LLM shows promise as an end-to-end therapeutic development assist, allowing one to query a single model for multiple steps of the development pipeline,” the authors noted in their paper.
The Broader Context
This advancement follows the release of Med-PaLM 2 in March of the previous year, which demonstrated an ability to generate more comprehensive medical answers than its predecessor, Med-PaLM.
The use of AI in drug discovery is rapidly expanding. Recent collaborations highlight this trend:
- In December, Absci partnered with AstraZeneca in a deal worth up to $247 million to accelerate novel cancer treatment discovery using generative AI.
- IBM and Boehringer Ingelheim announced a collaboration in November to explore the potential of generative AI and foundation models for biologic drug discovery.
- Other notable players include AI drug-discovery startups like Genesis, Daewoong Pharmaceutical, and AION Labs, which is an AI-driven partnership between global pharma and tech companies.
The Future of AI in Drug Discovery
Tx-LLM signifies a transformative step in the integration of AI into therapeutic development, offering a versatile and robust tool for streamlining multiple stages of the drug discovery pipeline. As AI capabilities continue to evolve, their application in healthcare and pharmaceutical industries promises to revolutionize the speed and accuracy of therapeutic innovations.
