SEATTLE– Synthesize Bio, a biotechnology startup pioneering biological foundation models to simulate gene expression experiments, has launched with $10 million in seed funding. The round was led by Madrona with participation from AI2 Incubator, Sahsen Ventures, Inner Loop Capital, and Point Field Partners.
The company has developed the Generate Expression Model-1 (GEM-1), a generative genomics foundation model trained on one of the most deeply curated RNA-seq datasets ever assembled. GEM-1 can generate in silico data that mirrors the results of laboratory experiments based on experimental design descriptions. According to the company, this breakthrough marks the emergence of “generative genomics,” where predictive models augment lab data, anticipate experimental outcomes, and ultimately accelerate therapeutic innovation.
“Biopharma companies need rich, representative data to identify new drug targets, validate safety and efficacy, and power foundation models,” said Rob Bradley, Ph.D., co-founder of Synthesize Bio. “Our models help scientists predict the results of new and currently impossible experiments, moving their science forward faster and more cost effectively.”
Synthesize Bio was founded by Bradley and Jeff Leek, Ph.D., both of Fred Hutchinson Cancer Center. Leek, Chief Data Officer at Fred Hutch and a Time AI 100 honoree, has led efforts to aggregate and normalize global RNA datasets into the largest combined dataset available. Bradley, Director of the Translational Data Science Integrated Research Center, has identified RNA dysregulation as a driver of cancer and contributed to the development of new cancer therapies.
“Madrona’s investment builds on our longstanding thesis at the intersection of AI and life sciences,” said Matt McIlwain, Managing Director at Madrona. “Synthesize Bio represents a transformative shift in how biopharma and researchers discover and develop new therapies. Rob, Jeff, and the team are uniquely equipped to bring generative genomics into practice, and we are thrilled to partner with them.”
The performance of GEM-1 was detailed in a recent bioRxiv preprint, showing its ability to predict the results of experiments conducted after its training cutoff and to generate synthetic data from large clinical cohorts. The company is working with biopharma partners to use GEM-1 in drug development, from identifying targets to simulating therapeutic responses and optimizing trial design. Access to GEM-1 is now available through Synthesize Bio’s platform and via R and Python API clients.
