About
Elizabeth Wood is the founder and CEO of JURA Bio, where she leads the development of AI-native drug discovery technologies that solve fundamental bottlenecks in biological machine learning.
JURA develops de novo AI models that generate the real-world data they need to learn. Under Wood's leadership, JURA has developed an integrated platform combining variational synthesis (1,000T+ real-world designs), VISTA screening (1B+ functional measurements per campaign), and LeaVS (codesign of experiments and learning for orders-of-magnitude information gain). This technology has enabled the discovery of millions of fully functional de novo therapeutic candidates across multiplexed therapeutic targets, validated CAR designs, and TCR-mimicking antibodies against previously hard-to-drug targets.
Background
Before founding JURA, Wood was a post-doc in the lab of Adam Cohen at Harvard's Department of Chemistry and Physics. She started her PhD with Angela Belcher and Markus Buehler at MIT, and finished it under the supervision of Claus Helix-Nielsen at the Technical University of Denmark, Departments of Physics and Civil and Environmental Engineering.
She has also worked at the University of Copenhagen's SBiN Lab with Kresten Lindorff-Larsen, integrating computational methods with experimental studies to better predict RNA structure and to understand how the ability of proteins to change their shape helps modulate their function.
Affiliations & Service
Wood is an affiliate member of the Pioneer Centre for Artificial Intelligence and serves on the board of Project Clio, a Massachusetts-based non-profit. She was previously a visiting scientist at the Broad Institute of Harvard and MIT, where she served on the founding steering committee of the Models, Inference & Algorithms (MIA) Initiative and as the primary organizer of the multi-year NeurIPS Workshop Learning Meaningful Representations of Life (LMRL).
Publications
Selected papers, preprints, and conference abstracts Google Scholar
Molecular machine learning
- Continuous Variational Synthesis Continuous-space training and post-training quantization make variational synthesis models easier to pre-train, fine-tune, and adapt to synthesis hardware.
- Manufacturing-aware generative models enable petascale synthesis of designed DNA Manufacturing-aware generative models turn controlled stochastic DNA synthesis into a physical sampling algorithm, enabling designed libraries at petascale.
- Accelerated Learning on Large Scale Screens using Generative Library Models Co-designs high-throughput screens and inference to learn from rare active sequences, using a generative library model to account for unmeasured negatives.
- Lifting Biomolecular Data Acquisition Extends compressed sensing to learned molecular activity functions, extracting more information from pooled high-throughput measurements.
- Estimating the Causal Effects of T Cell Receptors Uses naturally randomized T-cell receptor generation to estimate how individual receptor sequences affect patient outcomes from observational repertoire data.
Mathematical foundations
- Category Theoretic Analysis of Hierarchical Protein Materials and Social Networks Uses category-theoretic ologs to formalize structure–function relationships and draw precise analogies between protein materials and social networks.
- Maximum Entropy Distributions of Scale-Invariant Processes Derives a maximum-entropy description of scale-invariant natural processes from their scaling properties and macroscopic observables.
Conference abstracts
- Discovery and characterization of a first-in-class de novo AI-designed TCR-mimetic-CAR targeting HLA-A2:01+ PRAME+ cancers Conference abstract 293