Exclusive: Gen Z AI+Peptide Founder Lands Nearly RMB 50 Million Seed Round

Peptide drugs are undergoing a belated renaissance.

The explosive rise of GLP-1 drugs has shown the market that this "intermediate" molecule—sitting between small molecules and antibodies—combines the strengths of both: it can reach deep inside cells to target protein–protein interactions, act on cell-surface receptors, and even take on targets that have traditionally been considered "undruggable."

Meanwhile, the AI drug discovery paradigm is quietly entering its 3.0 era: the shift from AI-assisted drug discovery to AI-native drug design, with peptides as one of the core tracks in this leap. A startup less than three months old is trying to lead the next R&D paradigm shift in AI drug discovery through a peptide generative large model, breaking a long-standing industry pain point.

LigIT, an AI-native peptide drug developer, recently announced the completion of a nearly RMB 50 million seed round, jointly invested by Shanghai Xunfei Venture Capital Co., Ltd., L2F Lighthouse Founders' Fund, and a well-known industry player, with Lighthouse Capital serving as exclusive financial advisor. The proceeds will mainly be used for computing power leasing, core pipeline development, patent filings, talent recruitment, and daily operations.

Dr. Jia Yinjun, founder, CEO, and CTO of LigIT, was previously a postdoctoral researcher at the Institute for AI Industry Research (AIR) at Tsinghua University, and is the first-listed co-first author of the DrugCLIP study published in Science in early 2026. According to the paper, the method can screen hundreds of billions of molecules within one minute—a million times faster than conventional docking tools—truly making genome-scale virtual screening possible.

VCBeat sat down for an in-depth conversation with this Gen Z founder, seeking to trace a clear path: from clinical reflections in a PUMC ward, to technological breakthroughs in a Tsinghua lab, to industrial implementation at a startup—how Jia Yinjun and LigIT are turning a technological vision that once existed only in science fiction into real productivity for the pharmaceutical industry, step by step.

Jia Yinjun's own career path is itself a cross-disciplinary journey of convergence.

Admitted to the eight-year clinical medicine program at Peking Union Medical College, he began his studies on the clinical front line, yet ultimately turned to the very frontier of AI drug discovery.

"At PUMC you encounter many complex and rare diseases, and even the most outstanding clinicians can offer patients only very limited help," Jia recalls. "Back then I started to think: a doctor managing beds every day—how many people can he ultimately cure? If I could develop a new drug, perhaps it could benefit far more patients."

This plain, clinically rooted original intent planted the seed for Jia's later decision to move into basic research and drug development.

The release of AlphaFold2 in 2020 became a key turning point in Jia's career path. "It made me feel viscerally the disruptive power of AI for biomedicine. Starting from protein structure prediction, AI can amplify everyone's research capacity, and in the future it may help us understand more diseases and find corresponding solutions."

Earlier, during his undergraduate years, Jia had already built a technical foundation in computer vision and programming, and during his PhD he applied neural network methods to biomedical research—which enabled him to enter the field of AI for Science quickly.

He then joined the Institute for AI Industry Research (AIR) at Tsinghua University, working deeply in AI drug discovery and producing a series of major results: as first-listed co-first author, he published the DrugCLIP study in Science, completing the world's first genome-scale virtual screening and improving screening efficiency by a million times. In addition, he has published 9 SCI papers in top journals including Cell, and 10 papers at top AI conferences such as ICLR, ICML, and NeurIPS, accumulating a number of pioneering achievements in synthetic data, peptide generation, and other directions.

At a point when his academic results were breaking through one after another, Jia chose to leave academia and start a company full-time. In his view, "Drug development is an industrialized process. It needs a commercialization path and sufficient funding to truly turn AI capabilities into drugs patients can use. Entrepreneurship is a path closer to the 'curing the sick and saving lives' idea I had when I first studied medicine."

And choosing peptides as the core track for his startup, rather than small molecules where he had already produced top-journal results, also came from a clear industry judgment: "If I could choose only one drug modality and use a single AI model to cover as many disease targets as possible, peptides come closest to that goal. It's like a 'Swiss Army knife': small cyclic peptides can go deep inside cells to target intracellular targets, can act on extracellular targets, and can even take on some traditionally 'undruggable' targets—while usually offering safety advantages as well."

In fact, the team's exploration in peptides predates its work on small molecules. It was only that early public peptide data was extremely scarce, while the research foundation for small molecules was more complete, so they first produced results such as DrugCLIP in the small-molecule field; the two technical routes have always run in parallel. As synthetic data technology made breakthroughs, the peptide data bottleneck is expected to be unlocked, and Jia decided to make peptides the core direction of the startup.

And the seed of "solving human problems with technology," planted when he wrote the sci-fi novel "Renxin" in high school, has guided him ever since: "From imagining AI and the digitization of consciousness back then, to doing AI drug discovery now, my core idea has never changed—I hope to use technology so that everyone can have a healthier body and see a further future."

Supporting this ideal is a highly cross-disciplinary, composite team. 85% of LigIT's core R&D team hold bachelor's or doctoral degrees from Tsinghua University or Peking University, covering frontier fields such as AI algorithms, peptide chemistry, and drug development; the company has also invited several academic and industry authorities to serve as senior advisors. This complete "academia + industry + business" advisory matrix provides all-round support for LigIT's transition from technology to industry.

AI-driven peptide R&D has long faced three mountains: scarce structural data, models that are hard to scale, and the poor stability of natural peptides.

These three pain points have kept the industry stuck at the small-scale screening stage, making large-scale druggability hard to achieve. To address these underlying challenges, LigIT has built a three-tier, progressive technology system spanning data, models, and molecules, creating a complete foundation for AI-native peptide R&D.

The first tier is the data moat, centered on the PepUmi peptide synthesis dataset. Real 3D structural data in the peptide field is extremely scarce, far from enough to support generative model training.

Based on monomeric protein structures, LigIT extracts relatively independent, self-foldable loop regions from structures that do not have a bound peptide ligand, matches them with functional pockets within the protein, and then filters them through dozens of strict physical constraints to generate complex data that closely mimics peptide–target binding.

The PepUmi dataset built with this method has reached nearly 10B in data volume, making it one of the largest closed-source peptide synthesis datasets in publicly available information worldwide. More importantly, the team back-derives dataset bias through real wet-lab results and continuously optimizes data quality, so that synthetic data keeps approaching real biological scenarios.

The second tier is the model moat, centered on PepUni, a peptide generative large model. The scaling law is the core growth logic of the large model era, but in the peptide AI field, traditional models often embed a great deal of domain-specific inductive bias and have complex structures, making it hard to achieve linear capability gains through compute and data investment the way large language models do—which directly constrains the scaling of peptide AI models.

"When we were working on MEET at Tsinghua, one important idea was to keep the model structure as close as possible to the already mature language model and AIGC generative model systems," Jia explains. "That way we can reuse the mature experience accumulated over the past few years—model architectures, training frameworks, compute infrastructure—and make it easier for peptide generation models to complete scaling."

After LigIT was founded, it was the first to adopt an infra-friendly equivariant Transformer architecture for an AI peptide generative large model, producing its new-generation in-house model PepUni, which explicitly follows the scaling law: model capability can keep improving with data volume, parameter count, and compute investment. Experimental data shows that the PepUni series significantly outperforms existing public models on core metrics such as binding free energy and conformational confidence; the affinity of its generated cyclic peptides has reached the nanomolar level, truly achieving a scale leap in model capability.

The third tier is the molecule moat, centered on PepMirror, a mirror-image peptide generation model. Natural L-type peptides have poor stability and degrade easily—a long-standing pain point constraining peptide druggability; mirror-image peptides made of D-amino acids can be dozens of times more stable, but public data is scarce and traditional methods struggle to design them efficiently.

LigIT pioneered a zero-shot-generalizing mirror-image AI design approach: the model learns the structural correspondence between L- and D-amino acids and the rules of all-atom interactions. Experimental data on the D-type peptides it designs shows stability dozens of times that of conventional L-type peptides, with affinity at single-digit nanomolar levels for some targets—successfully achieving decoupled optimization of stability and affinity.

To bridge the "last mile" from computation to experiment, LigIT has also built a 72-hour peptide generation–synthesis closed-loop platform covering the full workflow of AI design, chemical synthesis, biological testing, and structure determination. The platform has now been validated on multiple oncology- and metabolism-related targets, fully verifying the real effectiveness of AI design.

On the business model, LigIT has chosen a "platform + pipeline" twin-engine path, balancing near-term technology monetization with long-term value growth, and progressively upgrading toward an AI-native pharmaceutical company.

In the near term, LigIT centers on platform technology assets, collaborating with pharmaceutical companies and research institutes through model licensing, API services, and joint R&D, to deliver AI-native hit molecules with high affinity and high stability and rapidly monetize its technology value.

To date, LigIT has established cooperation with multiple listed pharmaceutical companies, biotech companies, top-tier Class 3A hospitals, and national key laboratories in China, with several collaborative projects already entering the preclinical testing stage. Leveraging the advantages of purely AI-native design, its hit compounds show strengths on multiple druggability metrics and have won high recognition from partners.

In the long term, LigIT treats pipeline assets as the core carrier of value, independently advancing drug development for high-value targets, and amplifying value through subsequent BD deals, license-outs, and even independent clinical development.

"Our positioning has never been that of an AI technology services company, but of an AI-native pharmaceutical company," Jia states plainly. "For a pharmaceutical company, what matters most is always being able to deliver drug molecules that are truly usable by patients. Owning self-developed pipelines will definitely be our primary goal."

In pipeline layout, LigIT focuses on three major areas—oncology, autoimmune disease, and metabolism—with a selection logic that fits industry demand: PDC and ADC drugs in oncology, large unmet needs in autoimmune disease, and hot targets in metabolism are all high-value directions that MNCs and biotechs are paying close attention to.

At present, several LigIT pipelines are at the hit-to-lead stage, and by early 2027 the company expects to produce candidate compounds with druggability potential and solid experimental data support.

In Jia's view, "Peptide drug development as a whole is still at a relatively early stage, and there are still many high-value targets worth developing. In particular, breakthroughs in oral cyclic peptide technology since 2025 have opened up an entirely new market space—peptide drugs are no longer limited to injectable administration, and once extended to the oral setting, the boundaries of indications are greatly broadened: in theory, every monoclonal antibody drug could have an orally designed peptide counterpart, and behind that lies enormous commercial potential."

Looking at the global competitive landscape, overseas already has representative AI+peptide companies such as Vilya, incubated by a team led by Nobel laureate David Baker, with earlier technological starts; in China, the number of companies working on AI+peptide is also gradually increasing, but most still use a "traditional technology + AI assistance" model, and companies on a purely AI-native route remain relatively few.

Jia believes that for the peptide industry to move toward large-scale maturity, it still needs a revolutionary technology comparable to the humanized mouse platform in the antibody field—and AI is expected to be a core driver of that transformation. That is precisely where LigIT's long-term opportunity lies.

After this round of financing, LigIT will focus on three major milestones over the next one to two years: on the technology side, continuously increasing compute and data investment to further scale its in-house peptide generative large model and maintain global technological leadership; on the pipeline side, obtaining some IND-enabling molecules for partnered and in-house pipelines and advancing preclinical research; and on the commercialization side, seeking breakthrough progress in license-outs and exploring global collaboration opportunities.

For the longer term, Jia hopes that when the industry mentions LigIT, three keywords come to mind: AI-native, peptides, and global collaboration.

"AI-native is our core philosophy—we hope AI becomes the single most important force in drug design; peptides are our core track—we hope to become a leading player in it; and global collaboration is our corporate vision—we hope LigIT, rooted in China, starts from joint R&D and licensing partnerships and gradually covers various high-value therapeutic areas, growing into an AI-native pharmaceutical company with a global vision."

From clinical reflections in a PUMC ward, to technological breakthroughs in a Tsinghua lab, to industrial implementation at a startup, Jia Yinjun and LigIT are turning a technological vision from science fiction into real productivity for the pharmaceutical industry, step by step.

In the wave of AI reshaping pharmaceutical R&D, this young startup is using peptides as a fulcrum to pry open the paradigm leap of the AIDD 3.0 era—when AI is no longer merely an auxiliary tool but the core driver of drug design, the efficiency and boundaries of the entire new drug R&D process will be redefined.