Nurix and Loka Create DEL Foundation Model to Enable Proteome-Wide Screening for Drug Discovery
Industries
Life Sciences,
Biotech,
Healthcare, Therapeutics
Tech & TOOLS
AWS SageMaker, foundation models (LLMs), reinforcement learning, GenAI for molecular optimization
Teams & Services
Bioengineering,
ML, DevOps
milestones
• Unlocking drug targets once thought impossible
• Achieved significantly higher accuracy over previous models
• Screen unseen molecules against unseen targets
Industry
Life Sciences,
Biotech,
Healthcare, Therapeutics
Tech & TOOLS
AWS SageMaker, foundation models (LLMs), reinforcement learning, GenAI for molecular optimization
Teams & Services
Bioengineering,
ML, DevOps
milestones
• Unlocking drug targets once thought impossible
• Achieved significantly higher accuracy over previous models
• Screen unseen molecules against unseen targets
In the fight against disease, time matters. Nurix and Loka are harnessing AI’s most promising capabilities to shorten timelines and accelerate drug development.
The situation
Nurix, a biopharmaceutical company developing novel drugs for cancer and autoimmune disease, historically built a separate machine learning model for each discovery campaign — and each one needed its own program-specific data before it could contribute. Nurix's ML team saw that the scale and richness of their DNA-encoded library (DEL) could support something far more general: a single model that predicts against targets it has never seen.
Nurix had the science and the expertise. What they needed was the bandwidth and engineering capacity to build and deploy their vision at production scale. With infrastructure already running on AWS, AWS introduced Nurix to Loka, whose deep experience in generative AI and bio-focused foundation models made it a natural fit to build alongside them.
The challenge
Nurix's traditional drug discovery process required expert driven decisions across massive chemical and biological datasets. Loka partnered with Nurix to expand their advanced ML capabilities by building custom, domain-specific foundation models, integrating state-of-the-art approaches into DEL discovery workflows, and accelerating AI assisted data analysis, molecular interaction assessment and therapeutic candidate prioritization.
“Our DEL datasets have reached a scale and richness that make them ideal for advanced machine learning. The collaboration allowed us to unlock the predictive value embedded in this data.”
The solution
Together Nurix and Loka developed a foundation model trained on DEL data (DEL FM) to accelerate drug discovery and enable Nurix to conduct productive DEL screens in silico. Nurix set the scientific vision and the bar for success, guiding the model's development with domain and technical machine learning expertise. In close partnership, Loka brought its own ML expertise and the engineering muscle to translate that vision into a model deployed at scale on AWS.
DEL-FM combines state-of-the-art architectures, including protein and molecular foundation models, with Nurix's proprietary datasets. Unlike the legacy single-campaign models it replaces, DEL-FM learns from every past screening campaign to generalize across the proteome, extending robust predictions to new protein targets rather than one target at a time.
Loka brought specialized capabilities in generative AI, ML architecture design, and bioinformatics. Working alongside Nurix’s scientists and ML engineers, Loka helped build and optimize the DEL-FM framework, ensuring that the model could generalize across diverse protein classes and integrate smoothly into Nurix’s discovery workflows.
DEL-FM was trained, deployed, and operated on AWS SageMaker. Loka and Nurix built and optimized it together, with scientists from machine learning, computational chemistry, and the wet lab shaping a model that drops cleanly into Nurix's discovery workflows.
What we delivered
- DEL-FM, a foundation model trained on Nurix's proprietary DEL data. Once trained, it enabled in silico screening across billions of compounds and tens of thousands of proteins.
- A production-ready, scalable ML stack on AWS, built to support continuous model evolution.
Together, they enabled Nurix to screen unseen molecules against unseen targets, prioritize the right lab experiments earlier, and accelerate the discovery of new medicines from DEL data.
“DEL-FM extends the power of our DEL platform by predicting high-value binders across a wider range of protein surfaces, supporting the growing needs of induced proximity–based drug discovery. The collaboration with AWS and Loka made it possible to realize this capability far sooner, and it now plays an important role in how we initiate new programs while also significantly expanding the reach of our ligand discovery capabilities.”
Project results
Faster discovery. By inferring screening outcomes instead of running them, Nurix compresses timelines for early discovery.
New targets in reach. Nurix can now open discovery programs against targets once considered undruggable or those written off as unreachable by DEL due to experimental constraints.
Validated in the lab. Working from the model, Nurix has confirmed candidate molecules against several targets of interest — early steps toward new medicines.
