Beyond the Foundation Model
Why Biology Needs a New Kind of AI
At Ingenix, many of us came directly from the front lines of the enterprise AI revolution. In our previous work at Applica, we built the TILT model, one of the few AI models used by OpenAI as a state-of-the-art benchmark during the development of GPT-4. We had a close-up view of how enterprise AI evolved: from narrow automation tools into today’s versatile, LLM-powered platforms that can write code, analyze complex data, and support a broad range of knowledge workflows.
But when we turned our attention to drug development, it became clear that the rules were different. The approaches that reshaped enterprise AI simply don’t map onto biology. The biological world is too complex, too layered, and too context-dependent for anyone to build a single, all-encompassing “foundation model for biology.” Doing so would mean unifying the world’s dispersed biomedical data, mastering every biological scale, and investing staggering amounts of compute - an unrealistic proposition even for the largest organizations.
So we decided to take a different direction.
Rather than trying to construct one monolithic model, we developed a Model Fusion architecture. It brings together our proprietary models with the best open-source and domain-specific models, and integrates them into a system capable of simulating clinical drug safety and efficacy through step-wise, biology-grounded reasoning.
This is more than mere orchestration. The platform we’re building weaves together knowledge across molecular, cellular, tissue, organism, and population levels, and across a wide range of data modalities. Its purpose is not only to predict outcomes, but to explain them - to generate hypotheses that can be examined, challenged, and validated.
What has surprised us most is how consistently the fused models uncover mechanistic patterns and reasoning paths that no single model surfaces on its own. In our early case studies, the approach yields more interpretable insights and a workflow that feels closer to how scientists reason through complex biological problems.
As we begin this blog series, our aim is to share this journey in an open and thoughtful way. In the coming posts, we’ll explore the ideas, experiments, missteps, and breakthroughs that shape our work with design partners, and the questions that continue to push us forward.
If this topic resonates with you, we invite you to stay with us as the story unfolds. Each blog post will build on the last, gradually revealing how Model Fusion is taking shape and what it could mean for the future of clinical reasoning.
More to come soon.