Decoding the Language of Life: How AI-Driven Protein Design Is Reshaping the Future of Non-Addictive Analgesics

Take a common scenario in our daily lives as an example: parents, relatives, or friends endure the torment of chronic pain, which often recurs day after day and year after year. Many of the molecular targets underlying such pain may be intrinsically disordered proteins. In contrast to traditional analgesic treatments (such as morphine and fentanyl), although these agents are effective, they are associated with adverse effects including addiction, respiratory depression, and tolerance.Would it make a difference if artificial intelligence were involved in the development of analgesics? The answer is yes.As Artificial Intelligence Intervenes in the R&D of Analgesics, It Holds Promise to Reshape the Future of Pain Relief and Make the World a Better Place.

What Progress Has Artificial Intelligence Made in the Field of Protein Structure Prediction?Both the 2024 Nobel Prize in Physics and the Nobel Prize in Chemistry were awarded for research related to artificial intelligence. The most direct benchmark is the Nobel Prize itself: it typically takes decades for a scientific breakthrough to be recognized with this honor, yet protein structure prediction received the award just six years after its emergence. This rapid recognition underscores the explosive growth of the field and highlights the critical importance of studying protein structure and function for biologists.

The AI-driven protein drug design process begins with the identification of a functional site, which may serve as the therapeutic target (potentially an intrinsically disordered protein). This is followed by backbone design, sequence design, structure prediction, and druggability screening, culminating in experimental validation.Through this iterative process, it is expected to uncover what future analgesic drugs should look like.The protein backbone design model, trained by the team using big data and artificial intelligence network architectures, demonstrates superior performance, with certain metrics even surpassing those achieved by Nobel Prize-winning teams.

Photo of Professor Chen Haifeng's Lecture at the School of Life Sciences and Biotechnology, Shanghai Jiao Tong University

Scientific research is not just about data and models; the most moving aspects are often those related to people.During the pandemic, an undergraduate student proactively asked if he could undertake a research project related to artificial intelligence. I told him that it was certainly possible, encouraging him to give it a try without fear of making mistakes. He subsequently developed a protein sequence design model, which yielded surprisingly impressive results and laid a solid foundation for the later establishment of an AI-driven protein design platform. However, reality can sometimes be harsh—due to his insufficient academic ranking, he did not qualify for the postgraduate recommendation program at Shanghai Jiao Tong University, which I regretted at the time. Nevertheless, he went on to pursue a direct doctoral program at Peking University. This experience has increasingly reinforced my belief in one thing:Life offers many paths; you need not excel in all of them. By pursuing a single path to its utmost extreme, you will surely achieve success.

At the technical level, the team drew on the design principles of dual-target drugs such as tirzepatide,Developed a Dual-Target Protein Sequence Design ProjectTo evaluate the performance of multiple current artificial intelligence models, a “white paper” on the relevant algorithms was released in collaboration with the Nobel Prize-winning team, attracting widespread attention within the field.

The Sequence-Structure-Function Relationship of Proteins: Sequence Determines Structure, and Structure Determines Function. In research on intrinsically disordered proteins, traditional protein structure prediction tools such as AlphaFold primarily address the sequence-to-structure problem. However, a notable issue is:Our bodies contain a vast amount of proteins that lack defined structures; how, then, can we study their functions?This necessitates bridging the gap from sequence to function, constructing its structural system, investigating the distribution of dynamic structures, and performing dynamic structure prediction.

The team has developed relevant algorithms to address the direct generation of dynamic structures from sequences, enabling the study of function through these dynamic structures.Dynamic structures serve as therapeutic targets, laying a solid foundation for addressing currently intractable pain and other cancer-related issues. However, the repertoire of known drug targets remains highly limited. Future research must focus on another class of “dark matter”—intrinsically disordered protein (IDP) targets—by elucidating their dynamic structures and subsequently designing drug molecules that specifically target these dynamic conformations, ultimately aiming to resolve pain and a spectrum of related diseases.

To this end, the team developed IDPFold2, a dynamic structure prediction model for protein complexes based on a mixture-of-experts architecture.Compared with existing models, this model not only achieves higher accuracy but also better reflects the conformational distribution of intrinsically disordered proteins. Eighty percent of protein functions in the human body depend on protein–protein interactions, much like how social stability is maintained through communication and collaboration among individuals in society.Therefore, it is of great significance to investigate the structural changes of intrinsically disordered proteins during interactions.

In addition to artificial intelligence methods, to address the question of how intrinsically disordered proteins are distributed in the human body,Development of Accurate Ab Initio Molecular Force Fields via Physical Methods. Our team has been engaged in this work since 2013. At that time, during a presentation at Lanzhou University, colleagues from the Shanghai Institute of Materia Medica, Chinese Academy of Sciences, suggested that there were no robust research methodologies available in this field and proposed developing a novel approach to better study proteins lacking a fixed three-dimensional structure in the human body. This conversation was immensely helpful to me.Perhaps, in the process of communicating with others, certain perspectives and ideas may well become breakthroughs that open up new fields.It is precisely with such an accurate molecular force field, combined with backbone design, sequence generation, and dynamic structure prediction,We have developed a comprehensive, graphical, one-click general AI protein drug design platform, enabling efficient and accurate drug molecule design.This platform has been evaluated by China’s largest innovative pharmaceutical company as the nation’s premier algorithmic platform and has been used to support students’ scientific research competitions, consecutively winning the Grand Prize at two national “Challenge Cup” competitions.

How does this platform help us reshape the design of analgesics?There are approximately 1.5 billion chronic pain patients worldwide, with a market size exceeding $80 billion. As the national population ages, the prevalence of various chronic pain conditions increases with age, including neuropathic pain caused by cancer and advanced-stage diabetes. This demonstrates that the market both desires and requires the development of new drugs to address the health challenges facing humanity. Although many existing single-target opioid medications, such as morphine and fentanyl, are effective, they carry side effects such as addiction. In a collaborative research project with the Director of the Pain Department at Shanghai Tenth People’s Hospital and Xuzhou Medical University, one particular scenario left a deep impression. In the collaborating director’s outpatient clinic, large numbers of pain patients are seen every day. When physicians can offer little more than comfort due to the lack of effective medications, the sense of helplessness is very real. It is precisely because we have witnessed these realities that we have become even more resolute in our direction—To design a non-opioid receptor-targeting, non-addictive, and highly efficient dual-target drug.

“No Matter How Brilliant the Idea, Let the Data Speak.”We designed 10,000 sequences using the platform, selected 20 for chemical synthesis, and validated their efficacy and safety in two mouse models. The results demonstrated that the designed drug outperformed morphine in both neuropathic and inflammatory pain models, without adverse effects such as respiratory depression or constipation. The team further conducted experiments on addictiveness, toxicity, and drug stability. In the addictiveness study, mice in the morphine group exhibited significant drug dependence, whereas those in the drug group did not display similar behaviors. In the toxicity study, administration at ten times the effective dose for one week showed no significant differences in biochemical indicators compared to the saline control group, indicating a favorable safety profile. Drug stability tests revealed that 80% of the drug remained after four hours post-administration, demonstrating good stability.

The pain points of first-generation drugs are clear: their duration of action is too short, lasting only one hour.This means that patient compliance will be very low.To address this issue, we plan toLeverage algorithms to engineer long-acting drug molecules.In the inflammatory pain model, the modified drug took effect within 0.5 hours, with analgesic effects lasting for 6 hours, whereas morphine maintained its effect for only 1 hour. In the chronic pain model, the molecule coded as “66” demonstrated significantly superior efficacy compared to morphine in terms of acute, chronic, and long-acting analgesia. In the neuropathic pain model, the effective analgesic duration of this molecule extended up to 12 hours, with no significant difference from the saline control group observed until 24 hours. Given the faster metabolic rate in mice, if scaled proportionally to humans, its half-life is projected to reach 24 hours.

“What is the underlying mechanism?” This is a question every pharmacologist or researcher must ask themselves. The experimental results are before us, but what is the mechanism behind them? We must also conduct our own mechanistic validation.In terms of mechanism validation, biotin-labeled drug molecules were subjected to co-immunoprecipitation with two relevant targets, successfully detecting binding bands and confirming that the drug can target both key targets individually to exert its biological functions. Overall, the drug demonstrates good plasma stability and achieves long-acting effects after modification, exhibiting superior analgesic efficacy compared to morphine in both neuropathic and inflammatory pain models. Regarding safety, no respiratory depression, constipation, addiction, or toxic reactions were observed after multiple doses. Furthermore, there are currently no dual-target analgesic drug molecules on the market.Therefore, this is a first-in-class novel molecule that offers potent analgesia, broad coverage, non-addictive properties, no tolerance development, long-lasting effects, and a strong safety profile.

As we draw to a close, let us briefly review the research journey thus far.The Path of Scientific Research: The Value Lies in Perseverance and in Going with the Flow.Scientific research often requires long-term commitment. Taking the study of intrinsically disordered proteins as an example, initial efforts must focus on exploring novel methodologies. In the era of artificial intelligence (AI), it is essential to proactively integrate AI into research, making it a key driver for in-depth scientific advancement and clinical translation. With the support of Shanghai Jiao Tong University, the team has successfully spun off an AI-driven biopharmaceutical company, striving to accelerate the drug’s entry into clinical trials to effectively address chronic pain and improve patients’ quality of life.

All efforts converge on a shared vision: “A world free from the torment of chronic pain!”