NIH grant: SoS:BIO How Artificial Intelligence Shapes Science: Evidence from Structural Biology (1R01GM169340-01)

retained text

NIH grant 1R01GM169340-01: SoS:BIO How Artificial Intelligence Shapes Science: Evidence from Structural Biology
Institute: {'code': 'GM', 'abbreviation': 'NIGMS', 'name': 'National Institute of General Medical Sciences', 'admin_org_id': '12500024', 'admin_funding_url': ''}. PI: Ryan  Hill. Award: $206410.
Start: 2026-07-03T00:00:00.
PROJECT SUMMARY (See instructions):
Artificial intelligence (Al) is transforming scientific research, yet systematic evidence on how specific Al
tools alter the pace, direction, and productivity of science remains limited. This project investigates the
effects of AlphaFold2 - a deep learning system released by Google DeepMind in 2021 that predicts
protein three-dimensional structures with near-experimental accuracy - on structural biology and
downstream biomedical research. Understanding protein structure is foundational to medicine: it informs
how diseases arise and guides the design of drugs that bind to specific targets. Prior to AlphaFold2,
determining protein structures required expensive, time-consuming experiments; AlphaFold2 has since
made predicted structures available for over 200 million proteins, including the entire human proteome.
The project pursues three specific aims. Aim 1 examines how AlphaFold2 has affected structural biologists'
careers, including labor market outcomes for early-career researchers and salary and grant trajectories for
established scientists whose experimental expertise may be displaced. Aim 2 investigates how AlphaFold2
has altered the rate and direction of experimental structure determination, including whether it has shifted
research toward previously understudied proteins and changed individual researcher productivity. Aim 3
assesses downstream consequences for biomedical research and drug discovery, examining whether
predicted structures have expanded follow-on research on proteins lacking experimental counterparts and
accelerated early-stage pharmaceutical development.
The research design combines causal inference methods with administrative datasets - including the
Protein Data Bank, UniProt/Swiss-Prot, the Alpha Fold database, and OpenAlex - and an original survey
of active structural biologists. The core identification strategy exploits the universal release of AlphaFold2
predictions as a natural experiment, comparing proteins with and without prior experimental structures in a
difference-in-differences framework. The findings will contribute to the economics of science, the
economics of Al, and the empirical literature on how research tools shape biomedical discovery.

document fetched 2026-07-12 08:14