49 Billion Compounds, Nine Finalists, One Approval: Inside Mprosevir, China's First AI-Assisted Class 1 Drug

For years, the pitch for AI in drug discovery has been more promise than proof: dazzling platform demos, billions in venture funding, and a clinical pipeline that kept quietly shrinking. This month, that ledger finally has an entry on the other side. China’s National Medical Products Administration (NMPA) has granted conditional marketing approval to Mprosevir, an antiviral for mild to moderate COVID-19 in adults — and, according to Westlake University, the first AI-assisted “Class 1” innovative drug ever approved in China.

The approval matters on two separate axes at once. It is a milestone for AI-assisted drug development in the world’s second-largest pharma market, and it is a long-awaited validation of DNA-encoded library (DEL) technology, a screening approach first proposed 34 years ago that has spent its entire life searching for this exact moment.

What was approved

Mprosevir was jointly developed by Westlake University, Westlake Laboratory, and Westlake Pharmaceuticals in Hangzhou. Westlake University announced the approval in early September, noting that the program took just three and a half years from the initial discovery of the drug candidate to the completion of its clinical trials — a timeline that traditionally stretches toward a decade or more.

The “Class 1” designation, as Xinhua explains it, refers to innovative drugs that have never been marketed in China or abroad. It is the highest category in China’s drug registration system, and it demands three things: a completely new chemical structure, a novel mechanism of action, and clear clinical value. In other words, this is not a follow-on compound or a licensed import repackaged for the domestic market — it is original innovation from the ground up, aimed at SARS-CoV-2’s main protease (3CLpro), the same target family attacked by Paxlovid-style antivirals.

The NMPA’s grant is a conditional approval, a regulatory pathway China has used before for COVID-19 therapeutics, which lets a drug reach patients while confirmatory work continues. The indication is adults with mild to moderate COVID-19.

How AI actually earned its name in the project

The most interesting part of the Mprosevir story is not the approval itself but the workflow that produced it, because it shows AI playing a genuinely load-bearing role rather than a decorative one.

The Westlake team designed a dual-screening strategy that paired AI with DEL technology from the very outset. DNA-encoded libraries are a brute-force answer to chemistry’s search problem: every compound in the library is tagged with a short DNA sequence that acts like a barcode, letting researchers screen astronomically large chemical spaces in a single experiment instead of running millions of individual wet-lab assays.

The DEL library used to search for Mprosevir contained as many as 49 billion compounds. The DEL screen initially surfaced more than 100 “promising candidates.” Here is where the traditional process would have bogged down: testing each of those molecules individually — a typical next step — would have demanded substantial manpower, materials, and calendar time.

Instead, an AI model performed virtual screening on the candidate pool, eliminating false positives and identifying the molecules with genuine drug potential. The model narrowed more than 100 candidates down to just nine in a matter of days. When the team took those nine into activity testing, six demonstrated superior activity — a hit rate high enough that the R&D team later pointed to it as evidence that “genuine AI-assisted drug development is currently both efficient and feasible.”

After a series of toxicity, metabolism, and other drug-development assessments, the team selected the molecule WLU6937 as the lead candidate, which then became Mprosevir.

AI in the long middle, not just the first step

It is tempting to read this as a one-shot story — AI picks a molecule, humans do the rest. The details suggest otherwise. Westlake says AI “played a key role throughout” the following two years of optimization: enhancing the compound’s activity, improving its drug-like properties, reducing its toxicity, and advancing it from in vitro cell experiments and mouse studies through to clinical trials.

That matters because the industry’s hardest problem has never been finding a hit compound. It has been the unglamorous middle mile — balancing potency against toxicity, metabolism against manufacturability — where most candidates die. If AI is compressing that phase too, the 3.5-year timeline is not a curiosity; it is a preview.

The 34-year DEL milestone

For the DEL field, Mprosevir is arguably an even bigger deal than for AI. According to Tide News, Mprosevir is the world’s first small-molecule innovative drug successfully approved based on DNA-encoded library technology — a full 34 years after the concept was first proposed.

DEL has been popular in pharma for a decade; Amgen, AstraZeneca, and most large drugmakers maintain DEL platforms precisely because the technique can screen billions of compounds cheaply. But “used in discovery” and “produced an approved drug” are different achievements separated by an enormous valley of failed leads. The first DEL-originated approval gives the entire field a reference point that no amount of platform marketing could substitute for — proof that a barcode-tagged molecule can travel all the way from a 49-billion-compound library to a regulator’s stamp.

Why it happened in China

The Mprosevir approval lands amid a broader shift in China’s pharma industry. As Global Times notes in related coverage, heavy R&D investment, strong policy support, and global cooperation have driven rapid growth in China’s innovative drug sector — a reversal from the era when imported drugs were simply assumed to be superior to domestic ones.

Three structural factors line up in Westlake’s favor. First, China built a deliberate pipeline for AI-assisted drug discovery, with university-affiliated labs like Westlake Laboratory bridging academic research and commercialization vehicles like Westlake Pharmaceuticals. Second, the NMPA’s conditional approval pathway, battle-tested during COVID, gave regulators a familiar route to move quickly on an antiviral with unmet need. Third, the Class 1 classification creates a strong incentive to pursue genuine novelty rather than fast-follow licenses of Western drugs.

It is also worth noting the cautionary frame: a conditional approval for a COVID-19 antiviral in 2026 will live or die by its confirmatory data and real-world uptake, and a 3.5-year timeline for a target as well-studied as 3CLpro benefits from a decade of public pandemic-era science. Mprosevir proves the pipeline works; it does not yet prove it generalizes to harder targets.

What it means

For the AI-in-drug-discovery sector — which has spent years fielding questions about why no fully AI-native drug had crossed the finish line — Mprosevir is a concrete, citable counterexample of AI materially compressing the discovery-to-clinic journey inside a real regulatory process. It joins a small but growing set of proofs that machine learning can carry its weight at the candidate-selection stage, where its predictions are cheapest to test and most falsifiable.

For DEL practitioners, it removes the “no approved drugs yet” asterisk that has hung over the technology since 1992.

And for anyone watching the geography of biotech innovation, it is another data point that frontier drug development is no longer exclusively an American and European story. A university lab in Hangzhou just demonstrated, with a regulator’s signature, that AI-assisted original drug discovery can run end-to-end in China — on a timeline measured in years, not decades.