Merck Teams Up with Protillion as AI Protein Drug Discovery Moves Toward a Big Pharma Test

In new drug development, AI’s most compelling promise has never been to “replace” biology, but to help researchers see earlier which molecules are worth investing in costly and lengthy experiments. The drug discovery collaboration between Merck and Protillion Biosciences, based in Carlsbad, California, sits precisely at this critical point: using high-throughput protein data and machine learning to help identify new therapeutic candidates.

According to Genetic Engineering & Biotechnology News and company information released by Protillion on June 16, the agreement includes up to $510 million in potential milestone payments. Such amounts usually depend on whether later R&D, clinical, or commercial milestones are achieved, and are not the same as cash already paid; Protillion and Merck also did not disclose the disease areas, number of targets, size of upfront payment, or details of rights allocation in public materials.

Protillion’s focus is on quantitatively characterizing protein candidates at large scale, then enabling AI models to learn from those data. For antibodies or other protein drugs, for example, research teams need to know not only whether a molecule can bind to its target, but also to measure characteristics such as affinity, specificity, and manufacturability; these properties often determine whether a drug candidate can move from a screening list into a more rigorous development process.

This also makes the collaboration somewhat different from the usual “AI drug hunting” narrative. Protillion emphasizes on its website that the value of its platform comes from high-throughput, miniaturized wet-lab data, which it uses for model training and validation. In other words, AI here is not generating answers out of thin air, but is being placed into a loop that needs substantial measured protein data for support: design, measure, learn, and then return to select more promising molecules.

Merck’s role in the collaboration is to provide the drug discovery capabilities and downstream development judgment accumulated over many years by a large pharmaceutical company. For platform companies such as Protillion, being able to test their technology together with a multinational pharmaceutical company means their data generation and candidate screening capabilities will face standards closer to industry practice; for Merck, this is a way to bring external AI and protein engineering platforms into its early-stage R&D toolkit.

However, the publicly available information remains quite limited. The two parties have not yet provided model performance, dataset size, candidate validation results, or any preclinical success cases for outsiders to assess. For biomedical AI, the real threshold lies not only in algorithms, but in whether model predictions can be supported by reproducible experimental evidence and continue to hold up through toxicology, manufacturing processes, human trials, and regulatory review.

Therefore, this collaboration is best viewed as an industry test of an early-stage R&D platform, rather than a signal that a new drug is about to arrive. If Protillion’s platform can more quickly identify protein candidates with favorable properties in Merck’s projects, it will add a significant case for AI-assisted protein drug design; if progress is limited, it will also remind the market that the bottleneck in drug discovery is often not a lack of imagination, but a lack of molecules that can pass all the way through experimental and clinical testing.