September 30, 2026 — Southwest Research Institute (SwRI) is using a generative artificial intelligence (AI) tool to improve pharmaceutical development outcomes. The SwRI-developed GAMES, or Generative Approaches for Molecular Encodings, leverages large language models (LLMs) to produce text strings representing chemical structures.
The tool’s graphical user interface taps into the power of AI to make informed chemistry decisions at an accelerated pace. The model offers a complementary addition to SwRI’s robust computer-aided drug design platform Rhodium™.
“Drug development is a little like a marathon; in the beginning there are numerous candidates but for various reasons not all finish the race. Maybe a viable candidate isn’t stable enough, or it turns out to be toxic,” said Staff Scientist Dr. Jonathan Bohmann, one of the inventors of GAMES. “If during preclinical development we hit a roadblock, GAMES can identify a path forward.”
The ranking system embedded into GAMES points to the most promising and viable compound structures based on the properties of drugs already approved by the U.S. Food and Drug Administration. Where Rhodium serves as a drug design and screening tool, GAMES uses LLM and machine learning technology to generate models of viable compound structures trained to avoid issues with solubility, toxicity, and off-target effects and therefore more likely to achieve FDA approval. Drug development scientists at SwRI have already deployed GAMES to advance several drug development projects.
“GAMES generates new drug sequences from existing examples. This provides the model with implicit background that improves accuracy and consistency,” said Senior Research Engineer Jake Janssen, a member of the development team for GAMES.
For the next phase of research, the team relied on reinforcement learning from human feedback to improve the output of GAMES. SwRI deployed the toolkit to identify 18 antiviral candidates with improved likelihood of FDA approval in treating filovirus infections, such as Ebola. The method factored in chemical inventories in the domestic supply chain to procure precursors compatible with the synthetic route.
“Analysis of compound recommendations indicate that GAMES’ recommendations are in the top 10% of ranked candidates. Over a quarter were considered for follow-on studies and even reflected the latest work of chemists on the project,” said Research Engineer William Watson, a project team member.
GAMES was funded through SwRI’s Internal Research and Development Program, which provides resources for future-focused, unproven concepts to advance technology for government and industry clients. In fiscal year 2025, SwRI invested more than $13 million to fund IR&D projects to pioneer new technologies, expand institutional knowledge and capabilities, and encourage the professional growth of its staff.
Research Scientist Daniel Hinojosa will present “Generative Approaches for Molecular Encodings for AI-Enabled Medicinal Chemistry Workflows” at the Cancer Prevention and Research Conference in Houston, Texas, October 5-7.
For more information, visit Pharmaceutical Development and Machine Learning in Drug Discovery or contact Camaron Brooks, +1 210 522 2357, Communications Department, Southwest Research Institute, 6220 Culebra Road, San Antonio, TX 78238-5166.