Industrializing scientific discovery to radically improve lives.

Our Mission and Vision

In May 2023, Valence Discovery joined forces with Recursion to create Valence Labs: an AI research engine within Recursion dedicated to industrializing scientific discovery to radically improve lives. 

We are harnessing the power and promise of scaled compute and algorithmic innovation to solve fundamental challenges in biology, chemistry, physics, and machine learning. And we’re closely integrating these advances with Recursion’s capabilities in data generation and lab automation as the means through which we realize our impact. 

By acting boldly, leaning into risks, embracing failure, and trading incremental improvements for breakthrough advances that will redefine our field, we commit to realizing our ultimate vision: the development of highly autonomous systems capable of Nobel-worthy insights to unlock a new age of scientific discovery. 

Today, on the back of the exciting progress we’ve made with LOWE—an LLM-Orchestrated Workflow Engine for executing complex drug discovery workflows using natural language—we aim to share more about what we’re working on and why.

Our Strategy

Operating autonomously from Recursion, Valence Labs is a bold bet that the future of drug discovery—and indeed scientific discovery itself—will be driven by highly autonomous systems capable of reasoning about, learning from, and executing complex scientific tasks with minimal human input. 

Our strategy for building this future is straightforward: de-risk then integrate the individual components we view as necessary for autonomous scientific discovery:

  1. Foundation models that accurately represent or simulate the biological and chemical worlds of drug discovery.
  1. Inference engines and active learners to formulate hypotheses and learn from results.
  1. Orchestration and reasoning systems to design and execute optimized experiments for hypothesis testing. 

By combining the intellectual freedom of academia with the resources and stability of industry, we have a credible shot at making this vision a reality. Notably, we will leverage the full weight of the Recursion OS, which provides significant advantages over other labs pursuing similar challenges in the form of scaled data, compute, and lab integration:

  1. Data. Recursion has spent the past decade generating one of the world’s largest biological and chemical datasets. This allows our team to work not only on the same public domain datasets as our peers, but with massive-scale, fit-for-purpose, interoperable training data specific to our needs.
  1. Compute. Recursion will soon own and operate the most powerful supercomputer of any biopharma company (expected to be among the 50 most powerful supercomputers in the Top500 list) by adding >500 NVIDIA H100 GPUs to the >300 NVIDIA A100s already in place. In a world where scaling laws are driving unprecedented performance gains across nearly every application in AI, compute of this magnitude is a necessary prerequisite for unlocking the true power of Recursion’s data.
  2. Lab Integration. Unlike many of our peers, our team has the ability to bridge the gap between in silico hypothesis and in vitro and/or in vivo validation at unrivaled scale. We are constructing our research plans such that we take advantage of these virtuous cycles between atoms and bits. For example, we’re proactively pursuing problems where generating small amounts of fit-for-purpose training data can have an outsized ROI on model performance.

Our Progress

There is undoubtedly a lot of work to be done. But we’re proud of the progress we’ve made only ~6 months into our journey as Valence Labs: 

Research. Valence Labs continues to publish research papers at top-tier conferences in areas underpinning our strategy (such as causal representation learning, molecular representation learning, generative flow networks, generative chemistry, quantum mechanical simulations, and more). We aim to be open about the basic research we’re doing as we relentlessly pursue our mission. 

Open Science. Beyond publishing primary research, we strive to be leaders within the machine learning community through the open-science and open-source efforts Valence has become known for. These include: 

Autonomous Scientific Discovery. At the 2024 J.P. Morgan Healthcare Conference, we unveiled LOWE (our LLM-Orchestrated Workflow Engine), an LLM agent underpinning the Orchestration and Reasoning pillar of our strategy and the next evolution of the Recursion OS. LOWE democratizes access to cutting-edge computational tools by putting the power of Recursion’s immense datasets, foundation models, and fit-for-purpose computational tools in the hands of every Recursion scientist in a simple and scalable way.  

Consequential missions like ours require incredible focus, resources and, most importantly, brilliant scientists and engineers motivated by the desire to see AI make a positive and lasting impact on the world. Our vision for the future of scientific discovery is bold, but by building a team of passionate people capable of leveraging Valence Labs’ unique advantages, we are uniquely positioned to make our vision a reality. If you are inspired by our mission, we invite you to learn more about Valence Labs and apply to our open positions today!