Top-Tier Institutions Bet Big: AI Breaks New Ground in Drug Delivery

The hot air surrounding ITBT is still blowing, but its temperature has dropped significantly.

During the 2021 ITBT boom, AI drug discovery, as the most established and mature subsector, attracted significant attention. In the first generation of AI drug discovery companies, the focus was on the small-molecule drug development process; in the new wave of the past two years, large molecules have taken center stage with the rise of AlphaFold.How will AI continue to empower the development of more complex, next-generation therapeutics such as cell therapies, nucleic acid drugs, and gene therapies?

Drug delivery is an answer; it may not be standard, but it holds significant promise.

Since small molecules, AI-driven drug discovery has not seen any major opportunities for a long time.

It has been a year and a half since the last wave of excitement in AI-driven drug discovery. Following AlphaFold’s victory at the 14th Critical Assessment of Protein Structure Prediction (CASP) in December 2020, the field witnessed a surge of interest in macromolecule-based therapeutics.However, after a full year, AI-driven macromolecule development remains an early-stage sector, still far from practical application.

Even with AlphaFold2 being open-source, numerous challenges persist in the field of AI-driven macromolecular drug development.Ma Rui, Partner at FreeS Fund“It is believed that AlphaFold has accomplished a zero-to-one breakthrough, effectively solving the sequence-to-structure prediction problem. However, subsequent steps—particularly predicting function from known protein structures—have not been thoroughly investigated. A more ideal scenario would actually involve the reverse approach: given a specific target, directly designing a macromolecular sequence that binds to the target or achieves a desired biological function. These capabilities remain at some distance from practical application.”

For a variety of reasons, small-molecule drug development remains the only truly viable pillar of the AI-driven pharmaceutical industry to date. In many contexts, the two are even considered synonymous. However, even in the highly commercialized realm of AI-enabled small-molecule discovery, significant obstacles persist.

The most critical shortage for AI in small-molecule drug discovery remains data. Public datasets are the primary source of data for the AI-driven pharmaceutical industry. However, to obtain more first-hand information on drug R&D—particularly by integrating small-molecule data with pharmacokinetic, pharmacodynamic, and clinical data—the only viable approaches, apart from collaborating with pharmaceutical companies, are to establish proprietary laboratories.

Consequently, starting in the second half of 2021, the concept of “dry and wet labs” began to appear frequently in the primary market."Wet lab" refers to research conducted in a laboratory using molecular, cellular, and physiological experimental methods, which is characteristic of traditional drug discovery laboratories. In contrast, "dry lab" involves research through computer simulations and bioinformatics methods, which is what AI platforms do.

The integration of wet and dry laboratory platforms enables a synergistic workflow: the wet laboratory platform generates extensive experimental data from real-world environments, while the dry laboratory platform leverages molecular simulations and artificial intelligence models to predict the physicochemical and kinetic properties of drug candidates. Finally, machine learning bridges microscopic interactions with macroscopic physical phenomena, facilitating continuous algorithmic iteration based on data generated from both wet and dry laboratories.

The concept of integrated dry and wet laboratories is not new; many companies that are now industry leaders equipped themselves with wet laboratories to generate proprietary data in their early stages of development. Examples include XtalPi, which built its own large-scale laboratory in 2018, and BioMap, which established an integrated dry-wet experimental platform from its inception.

However, the sudden and frequent mention of this concept as a core competitive advantage for many startups precisely reflects the data scarcity dilemma plaguing the AI drug discovery industry.

Obstacles in data acquisition channels have meant that, to date, the business model of AI-driven drug discovery remains predominantly centered on providing CRO services. In September 2021, two drug candidates from IceStone received successive IND approvals in the United States, marking a significant milestone for China’s AI pharmaceutical companies. However, this achievement came after six years of effort by IceStone, failing to demonstrate the purported efficiency advantages of AI in drug R&D.

As small-molecule drug development hits a bottleneck, AI-driven pharmaceutical innovation needs a new direction—one with commercial potential and ample room for imagination. To identify such opportunities, it is essential to break free from the existing industry framework and examine the sector from a different perspective.

Beyond the framework of molecular design, drug delivery may represent a distinct opportunity in the field of AI-driven drug discovery.

Active ingredients and routes of administration are two key elements that constitute a drug. Molecular design has always focused on the study of active ingredients, while AI has long been absent from the field of route of administration, which is primarily determined by dosage form.

“Although AI-driven macromolecule development presents significantly greater algorithmic challenges, its core logic remains quite similar to that of small molecules. However, drug delivery and molecular design address problems in entirely different dimensions,” said Lai Caida, CEO of Metagenomi (Jitai Pharmaceutical).

Molecular design focuses on the binding process between a molecule and its target protein, including whether an optimal binding pocket has been identified and how the target protein is subsequently modified upon binding. In contrast, drug delivery concerns the binding interactions among a molecule and multiple other molecules, requiring an understanding of how the active ingredient interacts with all excipients to determine the final configuration formed.

Ma Rui told us, “Drug delivery can create some rather distinct opportunities. It is evident that domestic drug R&D is heavily concentrated in oncology and autoimmune diseases. In reality, many other disease areas—such as certain metabolic and neurological disorders—also represent markets exceeding $1 billion. In some of these fields, there are currently no approved drugs in China, whereas corresponding indications have already received regulatory approval in the United States. Therefore, by leveraging the 505(b)(2) pathway to first demonstrate clinical value in the U.S. and then introduce the products back to the Chinese market, we can address many urgent unmet clinical needs with the greatest speed.”

Stereotypical biases often lead people to believe that new drug development must focus solely on the innovation of drug molecules. In China, innovative drugs are frequently confined to innovations in active pharmaceutical ingredients; whereas in the United States, formulation innovation via the 505(b)(2) pathway represents a significant model of drug innovation, with the number of newly approved formulation products ranking second only to that of new molecular entities (NMEs) annually. In 2020, among the 111 New Drug Applications (NDAs) initially approved by the FDA’s Center for Drug Evaluation and Research (CDER), 42 were NMEs, while 26 were new formulation drugs.

Whether transitioning from injectables to oral formulations or converting conventional tablets into sustained- and controlled-release drugs, the core objectives of novel drug delivery systems are to reduce adverse effects, improve patient compliance, and enhance therapeutic efficacy. As a reward for innovation, these new formulation drugs can also enjoy 3–5 years of market exclusivity.

Currently, there are very few companies truly leveraging AI for dosage form innovation, with Metagenomi being the earliest such company in China.

MTG Therapeutics, founded in 2020, has its core products approaching the clinical stage. The company’s most advanced candidate, internally coded as MTS004, is designed to address complications associated with neurodegenerative diseases. In early 2022, the Center for Drug Evaluation (CDE) formally accepted the Investigational New Drug (IND) application for MTS004. If development proceeds smoothly, Phase III clinical trial recruitment is expected to commence by the end of the year.

Behind MTS004, Metagenomi has more than ten additional pipelines, including small-molecule drugs and mRNA-based nucleic acid therapeutics, with indications spanning multiple disease areas such as chronic kidney disease, diabetic neuropathic pain, fibromyalgia, condyloma acuminatum, and multiple myeloma.

It took Metagenomi just over a year to build more than ten complex drug pipelines. The high efficiency of AI-driven drug discovery is fully embodied in Metagenomi.

“We began building the platform in Hangzhou in June 2020, completed the establishment of our R&D platform and product development team by the end of 2020, and started laying out our product pipeline in 2021,” said Lai Caida.

In the future, greater opportunities in the pharmaceutical industry will lie in novel therapeutic modalities such as nucleic acid drugs and gene therapies. Drug delivery is an indispensable component of these emerging therapeutics.

“Novel therapeutic modalities are inherently well-suited to AI, as DNA and RNA sequences operate in an encoded format,” said Lai Caida.

Due to the instability of their active ingredients in vivo, novel therapeutic modalities require more complex delivery systems. Nucleic acid therapeutics commonly employ non-viral liposomes as delivery vectors, whereas gene therapy utilizes biological delivery methods such as adeno-associated virus (AAV) vectors.

Companies have already entered these two areas through AI-driven approaches. In addition to innovating dosage forms for small-molecule drugs, DTI Pharma is also conducting research on liposomal delivery; whereas in the United States,Dyno Therapeutics is designing AAV viral vectors using AI.

Whether for non-biological or biological delivery, current efforts are focused on two key aspects: tissue targeting of vectors and manufacturing.

On AI’s Approach to Solving Tissue Targeting Challenges, Lai Caida Explains Their Methodology“For instance, to achieve lung-targeted delivery, we leverage a high-throughput experimental platform to conduct screenings related to lung cells. We then perform extensive animal studies to assess biodistribution and enrichment patterns, thereby identifying novel lipid structures and optimal lipid composition ratios, which are further combined with other formulation components. The design space for liposomes is vast; our existing library contains numerous examples, with the scale of LNP variants approaching hundreds of millions. By continuously screening these cases and incorporating feedback from wet-lab experiments, machine learning models can infer which feature values contribute to the desired functionalities. Through this iterative cycle of dry- and wet-lab experiments, we enable continuous optimization.” According to Lai Caida, they have already achieved LNP delivery to the liver, lungs, spleen, muscles, blood vessels, as well as certain local and peripheral nerves.

In CMC development, machines require greater collaboration with humans.

“Computers may well design an exceptional lipid, but it can be difficult to synthesize in practice. Therefore, we incorporate a dedicated filter in our algorithm to screen out such molecules. After initial screening, some remaining lipids are synthesizable and possess best-in-class potential, yet they cannot be prioritized as the first choice due to current synthesis costs. In such cases, human judgment is required,” said Lai Caida.

Artificial intelligence is not intelligent from the outset. Every AI technology must undergo a learning process, a phase that the field of drug delivery is currently experiencing as well.

Lai Caida told us, “The initial iterative results were less than ideal. For instance, with lipid nanoparticles (LNPs), the AI-designed LNPs we developed in the first few months exhibited very low delivery efficacy, far below the gold standard. However, after a period of training, quantitative changes led to qualitative breakthroughs; we might suddenly identify a key breakthrough and discover an optimal combination capable of precisely delivering therapeutics to a target organ. Subsequently, we perform fine-tuning of LNP materials and formulations, among other tasks, ultimately achieving more efficient, safer, and more controllable synthesis and scale-up. The methodology aligns with our small-molecule formulation development platform, differing only in logic and application scenarios.”

Investors are always generous with their money when opportunities arise.

Metagenomi and Dyno have received equal capital recognition both domestically and internationally.MTI Pharma has just completed its latest two rounds of financing totaling $150 million. The rounds were led by PICC Capital and China Life Equity, with participation from existing and new investors including Sequoia China, 5Y Capital, CMB International, Lightspeed China Partners, Monolith, and FreeS Fund. Meanwhile, Dyno raised hundreds of millions of dollars in its Series A financing completed in May 2021, attracting renowned funds such as GV, while existing shareholders Polaris Partners and KdT Ventures also chose to continue their support.

Ma Rui from Jitai Pharma, who has been following the company since its angel round, told us: “I haven’t yet come across any other company that leverages AI for non-viral delivery in the same comprehensive way as Jitai.”

The emergence of companies such as Jitai and Dyno Therapeutics has opened a new window of opportunity for AI-driven drug discovery. To date, there have been too few participants in this sector to allow for an assessment from the perspective of capital cycles.

But now, with these two companies paving the way, more entrepreneurs are likely to enter this sector soon.

In China, nucleic acid drugs and gene therapy are focal points in the primary market, yet a generational gap persists compared to international advancements. At this juncture, if AI-driven drug discovery can collaborate with biotechnology firms to achieve breakthroughs in delivery systems, it will not only open up new application scenarios for AI pharmaceuticals but also provide domestic biotech companies with an opportunity to leapfrog their competitors.

Regarding the future development trends of AI-driven drug discovery, Lai Caida believes that the key lies in expanding beyond the small-molecule domain: “While there are already many players in the field of AI-enabled small-molecule drug discovery, AI involvement remains limited in emerging therapeutic modalities such as nucleic acid drugs, gene therapy, and cell therapy, leaving substantial room for future growth. To maintain a long-term competitive advantage in drug R&D, AI must keep pace with the evolution of drug modalities by deeply engaging in the development of novel therapeutics that target the encoding process from DNA to RNA to protein, thereby facilitating stable delivery of products to target organs and achieving precise targeting and precision medicine.”