Google creates Tx-LLM for drug discovery and therapeutic development

Google Research and Google DeepMind recently released a paper announcing the creation of a new LLM for drug discovery and therapeutic development dubbed Tx-LLM, fine-tuned from PaLM-2.

Tx-LLM utilizes the tech giant’s PaLM-2, its generative AI technology that uses Google’s LLMs to answer medical questions. 

The drug discovery-focused LLM was trained using 709 datasets to target 66 tasks across the various stages of drug discovery, including evaluating efficacy and safety, predicting targets, and predicting ease of manufacturing. 

The LLM constructs the Therapeutics instruction Tuning (TxT) collection by interleaving free-text instructions with representations of small molecules, such as SMILES strings for small molecules. 

SMILES, or Simplified Molecular Input Line Entry System, is a typographical method using printable characters that represent molecules and reactions. 

TxT was then used to prompt and fine-tune Tx-LLM, the therapeutics large language model, to solve classification, regression and generation tasks involved with drug discovery and therapeutic development.

To use TxT to predict drug synergy, the researchers used prompts composed of instructions, context and a question.

Tx-LLM performed above or near the state of the art (SOTA) models for 43 out of 66 tasks, and exceeded SOTA models on 22 tasks. 

“Interestingly, we find evidence of positive transfer between datasets with diverse drug types, as training on datasets including biological sequences improves performances on molecular datasets,” the authors wrote. 

“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 LARGER TREND

Med-PaLM 2 was released in March of last year, and was found to generate more comprehensive answers to medical questions than the tech giant’s original version, Med-PaLM. 

Artificial intelligence capabilities are increasingly being used in drug discovery.

In December, Absci, a startup focused on developing generative AI antibody discovery technology, announced a potential $247 million partnership with pharma giant AstraZeneca to expedite the discovery of novel cancer treatments using genAI technology.

A month before, IBM and Boehringer Ingelheim, a German pharma company, announced a collaboration to harness the power of genAI and foundation models to further biologic drug discovery.

Other companies in the space include California-based AI drug-discovery startup Genesis, publicly listed Daewoong Pharmaceutical and Israel-based AION Labs, an AI-enabled drug-discovery partnership between global pharma and tech companies.

The HIMSS AI in Healthcare Forum is scheduled to take place September 5-6 in Boston. Learn more and register.