Our Take
Merck is buying speed and data volume, not proven efficacy—the deal funds research toward unspecified therapies, with no independent validation of Prot-MaP's ability to deliver better candidates faster than incumbents.
Why it matters
Pharma is consolidating around high-throughput experimental AI rather than pure computation. Merck's pipeline faces patent cliffs (Keytruda loses exclusivity soon), making faster antibody discovery operationally urgent.
Do this week
Drug discovery teams: audit whether your platform generates functional experimental data *before* feeding it to ML models, not after—Protillion's advantage rests on that inversion.
Merck signs multi-target deal with Protillion
Merck & Co. will partner with Protillion Biosciences on a discovery collaboration that could generate up to $510 million in milestone payments tied to research, development, and commercial success toward an unspecified number of therapies. The initial focus is inflammatory diseases, where Merck sees unmet need and differentiation opportunity.
Protillion's core technology, Prot-MaP (Protein Display on a Massively Parallel Array), combines high-throughput protein testing with proprietary machine learning. The platform uses Illumina DNA sequencing flow cells to generate tens of millions of immobilized proteins in a single run, enabling rapid testing of millions of therapeutic antibody variants simultaneously. Results arrive in as little as 48 hours (company-reported), compared to months using traditional methods.
The platform characterizes protein libraries at scale to avoid model overfitting, a common failure mode in AI-driven protein design. It can also engineer antibodies with difficult-to-achieve profiles—pH-dependent activation, multi-target specificity—by generating megascale quantitative datasets for protein design AI to learn from.
Protillion was founded in 2019 by Curtis Layton (CEO) and Will Greenleaf, a Stanford genetics professor. Layton developed Prot-MaP in Greenleaf's lab after completing a postdoc there. The company has 30 employees and is backed by investors including Illumina Ventures and ARCH Venture Partners. In March, it hired Robert Hollingsworth as Chief Scientific Officer; Hollingsworth spent 30+ years in biopharma, including roles at Pfizer, GSK, and MedImmune.
Merck is racing to replace blockbuster revenue
This deal sits within a broader Merck strategy to shore up its oncology and immunology pipelines. The company faces billion-dollar revenue losses as patent exclusivity expires for Keytruda (pembrolizumab, cancer immunotherapy) and Gardasil 9 (HPV vaccine). In recent months, Merck has inked collaborations with Quotient Therapeutics (up to $2.2 billion for somatic genomics in IBD discovery), Infinimmune (up to $838 million for antibody discovery), Google Cloud (up to $1 billion for agentic AI across R&D and manufacturing), Tempus AI (undisclosed, precision medicine biomarkers), and Mayo Clinic (undisclosed, virtual cell technologies).
The Protillion deal reflects a structural shift in drug discovery: pharma is acquiring experimental data generation capacity, not just computational models. Hollingsworth stated that Protillion takes the opposite approach from most competitors, which "start with AI and then look for data." Protillion generates the data first at massive scale, then trains AI on it. The bet is that high-quality, high-volume functional data beats limited datasets fed into sophisticated models.
What remains unverified
No independent benchmark compares Prot-MaP's discovery speed or candidate quality to existing platforms (phage display, yeast display, cell-based systems). Merck's milestone structure ties payouts to research, development, and commercial success—but those milestones, the number of programs, and the actual therapeutic candidates are undisclosed. The first two programs target inflammatory disease, which is Merck's strength in immunology; success there does not necessarily predict performance in oncology, neuroscience, or other areas Protillion claims the platform can address.
Hollingsworth claims the platform enables "unprecedented insight and precision" and can generate "tens of millions of clusters of immobilized proteins" in one run. These statements come from company personnel, not independent reproducers. Until a peer-reviewed study or third-party validation compares Prot-MaP candidates to benchmark therapeutics in side-by-side functional assays, the speed and quality gains remain vendor-reported.
Merck is investing $510 million in the possibility that data-first, experiment-centric AI beats model-first approaches. That possibility is plausible but not yet proven in the field.