Artificial intelligence (AI) is already reconstructing how scientists discovered and discovered new drugs. From predicting protein structures to optimizing clinical trials, AI tools can analyze biomedical data at speeds and measures that were unimaginable 10 years ago. However, as the White House's “American AI Action Plan” recognizes, it recognizes that reaching the full potential of AI in drug development requires significant upgrades in the way science is implemented and the infrastructure that supports it.
Building a Lab Infrastructure
The plan's appeal to “investment in AI-enabled science” reflects basic reality. Even the most advanced AI models are as useful as the scientific workflows they serve. In biopharma, AI can generate promising hypotheses, such as identifying new drug candidates for rare cancers. However, bottlenecks often occur in labs where traditional experiments are slow, manual, and resource-intensive.
An automated, cloud-enabled lab could change it, as the plan suggests. For example, AI-driven high-throughput screening can increase the efficiency, accuracy, and cost-effectiveness of drug screening and significantly accelerate drug development.
Supporting new scientific organizations
The plan also calls for the support of intensive research institutions (FROs). It is a nonprofit organization designed to tackle large, complex research issues such as tools and datasets and produce public goods. Previous examples include the large-scale hadron colider and human genome projects. These were efforts that went far beyond the capabilities of a single academic lab, company, or informal consortium, and were not profitable enough for the industry to take on.
At Biopharma, FROS was able to tackle the pre-competitive challenges for one company. One current example supported by FROS incubator Convergent Research aims to identify all unintended targets of approved small molecule drugs. Official benefits include excellent safety pharmacological profiling, opportunities for reuse of new drugs, and improved training data for AI drug development models.
Enhance your data infrastructure
Another priority in the plan is to “build a world-class scientific data set” while maintaining privacy, recognizing high-quality data as a national strategic asset, encouraging researcher data sharing, and establishing a secure computing environment for restricted federal data.
Privacy Enhanced Technology (Pets) is worth supporting and enabling more large-scale data sharing, including privacy differences, federated learning, secure multi-party calculations, and full homomorphic encryption. These tools allow sensitive data to be accessed, shared, and analyzed using advanced mathematical and statistical principles, without the disclosure of personal or unique information, in order to minimize the risk of re-identification or data leakage.
With AI-enabled drug discovery, pets help to overcome legal and institutional barriers that often prevent genome, clinical, and drug pooling. Public-Private Partnerships (PPPs) provide a framework for effectively applying pets, especially in pre-competitive spaces. The Innovative Health Initiative (IHI), Europe's largest biomedical PPP, was launched in 2008 by the European Union and the Federation of the European Pharmaceutical Industry Association (EFPIA). The melody project used federal learning to enable 10 pharmaceutical companies to collaborate on AI models for screening drug candidates, while keeping their own data secret.
In biopharma, data quality, diversity, and accessibility determine the predictive ability of specific AIs in patient subgroups that may benefit from protein structure modeling, simulating drug target interactions, or treatment. For example, Protein Data Bank was an important resource for AI-driven protein structure prediction via Alphafold. These resources can be expanded to open up new treatment opportunities by covering more diverse data types, including imaging and electronic health records (EHR).
Take home
AI can accelerate drug discovery and development, but without infrastructure, datasets and organizational models, many of the possibilities remain unrealized. The AI Action Plan's prioritization of “Investing in AI-enabled science” and “Building world-class science datasets” (automated cloud-based lab funding, FRO support, building high-quality datasets) could help reduce bench-to-bedside time and save US leadership in biopharmaceutical innovations.
Continuous public funding for basic science remains a critical gap. From mRNA vaccines to CRISPR gene editing, many of America's most transformative biomedical breakthroughs have emerged from decades of federally funded high-risk research, without immediate commercial application. Continuing investment in institutions such as the National Science Foundation and the National Institutes of Health is essential to drive fundamental AI applications for drug discovery, promote private sector investment, promote risky early stage innovation, and ensure that AI tools address public health needs. Without it, the US risks concessions to rivals that combine AI ambitions with robust investments in basic research.
