Merck and biotech startup Protillion Biosciences have launched a collaboration that could generate up to $510 million in milestone payments, combining artificial intelligence with continuous experimental feedback to discover new therapeutic antibodies faster than traditional methods. The partnership represents a shift in how major pharmaceutical companies are approaching drug discovery, moving beyond pure computational prediction toward a hybrid model that feeds real laboratory results back into AI systems.

What Makes Protillion's Approach Different from Other AI Drug Discovery Platforms?

Most AI drug discovery companies start with computational models and then search for data to validate them. Protillion took the opposite approach. The company built its Prot-MaP platform, short for Protein Display on a Massively Parallel Array, to generate enormous amounts of experimental data first, then apply AI to understand what makes the best candidates work.

Prot-MaP works by testing up to one million protein variants simultaneously in a single experiment and delivering results in as little as 48 hours. The platform generates tens of millions of clusters of immobilized proteins directly on an Illumina DNA sequencing flow cell through a process called tethered in situ transcription and translation. This allows researchers to characterize millions of antibody variants per run while avoiding the common pitfall of AI models overfitting to limited datasets.

"Many companies start with AI and then look for data. We took the opposite approach," explained Robert Hollingsworth, Chief Scientific Officer at Protillion. "Rather than relying primarily on computer predictions of protein structure, we can apply Prot-MaP to directly generate large-scale functional data and identify the best therapeutic candidates based on real-world experimental results."

Robert Hollingsworth, Chief Scientific Officer at Protillion Biosciences

The platform's strength lies in its combination of scale, speed, and machine learning integration. By generating millions of protein measurements in parallel, Protillion creates the kind of rich, large-scale datasets needed to train more powerful and predictive machine learning models. Those models then help design and optimize better therapeutic candidates faster and with greater precision.

How Does This Collaboration Address Merck's Drug Pipeline Challenges?

Merck faces significant pressure to replenish its pipeline as patent exclusivity expires for blockbuster drugs like Keytruda (pembrolizumab), a cancer immunotherapy, and Gardasil 9, a vaccine. The company has launched several technology-focused collaborations in recent months to develop new therapies in cancer and immunology. This partnership with Protillion is part of that broader strategy.

The initial collaboration will focus on two programs targeting inflammatory diseases, an area where Merck sees significant unmet medical need and strong opportunities for differentiation. However, the Prot-MaP platform's capabilities extend far beyond inflammation, enabling the discovery and development of novel biologics across a broad range of therapeutic areas including oncology, vaccines, infectious diseases, cardiometabolic and respiratory diseases, neuroscience, and ophthalmology.

How to Evaluate AI-Powered Drug Discovery Partnerships

• Data Generation Capacity: Look for platforms that can generate millions of experimental data points rapidly, not just rely on computational predictions alone, as this reduces the risk of AI models making inaccurate predictions based on limited training data.
• Real-World Experimental Validation: Assess whether the platform feeds actual laboratory results back into AI systems continuously, creating a feedback loop that improves predictions over time rather than using static datasets.
• Therapeutic Profile Complexity: Evaluate whether the platform can engineer biologics with sophisticated characteristics such as pH-dependent activation or multi-target specificity, features that are difficult to achieve with traditional methods.

The Merck-Protillion deal structure reflects confidence in this approach. The collaboration includes a multi-target discovery and license agreement, with Merck gaining access to Protillion's platform for discovering novel therapeutics while Protillion receives milestone payments tied to the successful development of drug candidates. This arrangement aligns both companies' incentives around actual therapeutic progress rather than just technology licensing.

"Prot-MaP is a technology platform that allows us to test millions of protein interactions simultaneously, generating an unprecedented amount of data in a matter of days rather than months. This gives us the ability to engineer antibodies with highly specific characteristics, such as stronger and more precise target binding, the ability to engage multiple targets, or the ability to activate only under certain physiological conditions," said Hollingsworth.

Robert Hollingsworth, Chief Scientific Officer at Protillion Biosciences

The platform was invented by Protillion CEO and co-founder Curtis Layton and co-founder Will Greenleaf, a professor of genetics at Stanford University School of Medicine. Layton developed Prot-MaP while working as a postdoctoral fellow in Greenleaf's lab at Stanford, then founded Protillion in 2019 to commercialize the technology. His work pioneered a new approach to high-throughput interrogation of biochemical systems by uniting protein engineering, next-generation sequencing technology, molecular biology, in vitro transcription and translation, computational biology, and software development.

This partnership signals a broader shift in pharmaceutical innovation. Rather than waiting for AI to mature in isolation, major drug companies are now integrating AI into their discovery workflows in ways that combine computational power with experimental reality. The $510 million in potential milestone payments suggests Merck believes this hybrid approach could accelerate the discovery of therapeutically meaningful antibodies across multiple disease areas, potentially shortening timelines from years to months for early-stage candidate identification.