Over the past 20 years since the merger of 芭乐视频 and 芭乐视频 Second Medical University, the University has established dedicated funding for medicine–engineering integration. By continuously building platforms and exploring new mechanisms, SJTU has enabled deeper collaboration between medicine and engineering, producing numerous nationally recognized research achievements and providing strong support for the University’s overall leap in strength and disciplinary development.
To reflect the exploration and achievements across the University in this field, SJTU News will publish a series of exemplary cases. The series aims to review the development path, distill practical experience, and further encourage faculty members and medical professionals to commit to interdisciplinary integration—advancing scientific innovation and medical progress, and contributing to the Healthy China strategy.
Hong Liang holds a PhD in Science and is a professor. He is currently a specially appointed professor at 芭乐视频 across the Institute of Natural Sciences, the School of Physics and Astronomy, the School of Pharmacy, and the Zhangjiang Institute for Advanced Study. He began his undergraduate studies in the Department of Physics at the University of Science and Technology of China in 2000 and earned his bachelor’s degree in 2004. After completing a master’s degree in physics at The Chinese University of Hong Kong, he pursued doctoral research in polymer science at The University of Akron in the United States under the supervision of Alexei P. Sokolov.
In 2010, he joined Oak Ridge National Laboratory as a postdoctoral researcher, focusing on computing and artificial intelligence under Jeremy C. Smith. He joined 芭乐视频 in December 2014. His main research interests include using AI to design proteins and drug-like molecules with specific functions, investigating protein structure, dynamics, and function, and developing cryopreservation technologies for biomacromolecules.
He has received multiple research grants, including Key and General Programs from the National Natural Science Foundation of China, as well as major projects funded by the 芭乐视频 Science and Technology Commission. He has published more than 80 SCI-indexed papers. In recent years, he has produced pioneering results in AI-driven protein functional design and applied these advances to the development of high-performance proteins, empowering scientific research and product R&D for more than 20 research institutes nationwide and over 30 companies in China and abroad.
In many ways, the forms life takes are, at their core, expressions of protein function. Today, people often talk about how to consume or supplement protein “scientifically” for better health or faster recovery, yet the fundamental scientific questions around proteins have challenged researchers for more than half a century.
In 2024, the DeepMind team was awarded the Nobel Prize in Chemistry, and AlphaFold2—developed by the team—was recognized for solving, for the first time, the problem of predicting 3D structure from a protein sequence. In the post-AlphaFold era, the central question in protein science has shifted even more toward function: only proteins with strong functional performance—high activity, high selectivity, and high stability—can become truly commercial protein products.
Hong Liang saw early how this key challenge could be paired with AI. Riding the convergence of artificial intelligence and life sciences, he and his team have fused the rigor of physics, the sophistication of polymer science, and the creativity of AI into a toolkit for decoding life’s molecular “language.” Through interdisciplinary thinking and engineering-oriented practice, they are pushing AI for Science beyond theory and toward industrial application.
Hong Liang’s team at SJTU’s Institute of Natural Sciences
Three major pivots—driven by a simple goal: “I wanted to build something real.”
Hong Liang began his research journey in physics, but he has never allowed disciplinary boundaries to define what he can or cannot do. In his view, physics is fundamental—less about producing a specific “deliverable,” and more about a way of thinking. Band theory, for example, emerged from physics, yet it was researchers in microelectronics who ultimately built semiconductors; quantum computing has followed a similar pattern. The same goes for superconducting materials: physicists proposed the theories, but turning them into real materials largely fell to materials scientists.
That simplest motivation—“I wanted to build something real”—is what led him, after finishing his undergraduate degree, to move to Hong Kong to study materials physics and chemistry, marking the first cross-disciplinary step of his academic path.
At The Chinese University of Hong Kong, he began working with nanomaterials, focusing primarily on thin films and nanowires in semiconductor systems. For his PhD, Hong made another deliberate pivot and went to The University of Akron for polymer research. The campus sits near two globally well-known tire and rubber companies—Goodyear and Bridgestone. Industry demand helped shape the university’s research strengths and reputation.
But to Hong, the conventional polymer materials track felt short of novelty. As he put it, tire research often looked like “cut the rubber, add something, stretch it—and repeat.” He therefore made a decisive switch to biopolymers and proteins, focusing on the physicochemical properties, dynamics, and phase transitions of polymers and proteins.
After completing his PhD, Hong became increasingly aware of the limits of traditional experimental methods—especially the lack of single-molecule techniques, which constrained the study of microscopic mechanisms.
He recalled: “In my last year of PhD, I traveled across the U.S. to attend talks. I saw people using computational simulations to study interactions between small molecules and proteins, and I found it fascinating.” That experience drew him toward computation, leading him in 2010 to begin postdoctoral work in computational biology at Oak Ridge National Laboratory.With the emergence of AlphaFold, Hong realized that traditional physics-based computation was often used mainly to interpret experimental results after the fact, whereas AlphaFold could provide prior predictions—at a level of accuracy far beyond conventional physics-based methods. That shift ultimately led him to move decisively into AI.
When talking about AI, Hong jokes that he “earned an AI degree on Bilibili.” He entered the field through Professor Hung-Yi Lee’s AI courses posted on the platform. After 80 full lectures, he developed a much clearer understanding of AI—and became even more committed to the path of AI for Science.
From 3D structure to a biological language: AI as a new key to decoding proteins
In the post-AlphaFold era, the focus has indeed shifted toward protein function—but predicting function is extremely hard. A protein sequence may change by as little as 1%, yet the resulting protein can lose 95% of its activity, or even lose biological function entirely. And when those altered sequences are fed into AlphaFold2, the predicted structures often look almost unchanged.
This highlights a crucial point: structure is not the same as function. Structure may be necessary for function, but it is far from sufficient—often very far. In protein engineering, therefore, one cannot look at 3D structure alone; instead, the amino-acid sequence should be treated more like a symbolic “biological language.”
Hong explained: “You can think of an amino-acid sequence as a piece of text. If you look at nature, statistical analyses show that the total number of complete protein sequences discovered across humans, animals, insects, bacteria, archaea, and other organisms is still under one billion—roughly 10^9.
But how many sequences are possible? Suppose a protein is 400 amino acids long, and each position can be one of 20 amino-acid types. Then the number of possible combinations is 20^400—far, far larger than 10^9. In other words, after natural selection and evolution, amino-acid arrangements in real proteins are extremely constrained.
If we can learn the rules behind those constraints, we can design high-quality proteins that follow nature’s logic—much like ChatGPT learns the rules of language and then generates coherent text. This kind of grammar- and rule-based design is only feasible with large models and substantial computing power.”
“Technology shifts in an industry don’t bend to individual will,” Hong said. Facing an unstoppable wave of AI-driven change, he began as early as 2020 to combine AI, computation, and wet-lab experiments for protein design.
Over the past few years, the cross-disciplinary team he organized has continuously worked on data collection, cleaning, labeling, and training deep learning models, ultimately building the Venus series—general-purpose large models for protein design. Some models focus on improving catalytic activity, others on thermal stability, others on tolerating extreme pH, and some even enable synthesis for non-natural substrates. Overall, the Venus series has strong general capabilities, and certain performance metrics have surpassed comparable products that leading global biotech companies had dominated the market with for more than a decade.
Hong Liang (right) discussing with colleagues
Looking back on the process of applying large AI models to protein engineering, Hong admits he faced many challenges. “The hardest part was the early stage—experimental validation and getting people to try it.” Scientific work must prove its value through experiments; it cannot stop at building theoretical models.
When he first built the models and showed them to friends and companies, the algorithms looked impressive, but no one was willing to take the risk of trying them. To break the deadlock, Hong decided to validate the approach himself. He published related work in late 2022 and began running wet-lab verification. After two rounds of experiments, the results were encouraging.
He then shared the findings with experts and scholars—including faculty such as Yang Guangyu and Feng Yan from SJTU’s School of Life Sciences and Biotechnology—as well as with industry partners. He also began engaging directly with companies, using his own algorithms to meet concrete industrial needs and推动 specific industrialization projects.
Beyond publishing in top journals: building products industry will adopt—and that can truly land in practice
For Hong, the ultimate meaning of “building something real” is practical use—turning research into competitive products. In today’s environment, he argues, publishing papers is no longer the only goal for researchers; solving real engineering problems matters even more.
“We’ve published many strong papers, and a lot of that work is genuinely useful—industry recognizes it,” he said. “China is facing intense international competition. If we remain at the stage of only publishing good papers, then in 10 or 20 years our technology may truly fall behind. We have to contribute to national technological competitiveness and productivity.”
Take the Venus series as an example: it designs protein sequences directly around functional requirements. The team has already delivered what it describes as the world’s first and second high-difficulty protein products designed by large models and successfully industrialized.
After proving that a general-purpose AI system for protein engineering could truly work end-to-end, Hong’s team worked hands-on with more than 30 companies, helping them develop protein products. Seeing molecules designed on a computer being produced in 5,000-liter fermentation tanks—and then used in real life—was, in his words, “an overwhelming sense of happiness for someone who once did purely basic research.”
Hong Liang (second from right) at a 5,000-liter scale production site
Beyond protein products, Hong’s team has also achieved multiple breakthroughs with collaborators in adjacent cross-disciplinary areas such as new energy, biopharmaceuticals, and breeding. This interdisciplinary way of thinking grew naturally from Hong’s own path—and has become a defining feature of his research career.For example, he led a collaboration with Professor Xu Ping from SJTU’s School of Life Sciences and Biotechnology to convert carbon dioxide into green energy—an approach that could help reduce dependence on petroleum.
In another industry collaboration with the in-vitro diagnostics company Zhongyuan Huiji, the team co-developed a high-activity immunodiagnostic labeling enzyme—alkaline phosphatase. Compared with products from leading international companies, its activity has reportedly reached three times higher, showing promise for applications such as Alzheimer’s disease testing. The project has already completed 200-liter scale-up production.
All of these projects are ultimately driven by real problems. And to find real problems, Hong believes, you must understand market demand and step outside the lab to build relationships with industry.
“You have to convince your users,” he said. “As an engineer or technologist, you need people to genuinely want to use your product—and to believe it’s worth a company’s investment.” To make that happen, substantial preparation is required well in advance.Hong emphasized that the most important shift is mindset: “Researchers in basic science have to look at their work through the lens of industry–academia collaboration, step out of their comfort zones, and test whether industry will accept it.”
“Even research like AlphaFold may not directly produce a product,” he added, “but countless people are using it today. That success also came from stepping outside the comfort zone.”Today, Hong’s team has partnered with more than 30 companies and over 20 universities. “Now I don’t really need to promote our large-model design capability anymore—people come to us,” he said.
Keeping pace with rapid technological change while remaining grounded in practical work, Hong has led his team to take experimental results out of the lab and into companies—ultimately pushing their products into real-world use for a broad base of end users.
Hong Liang speaking at SJTU Summer School on “AI in Everyday Life and Science”
(Adapted from Interdisciplinary Integration: Medicine & Engineering in Action—Twenty Years of Medicine–Engineering Integration at 芭乐视频, 芭乐视频 Press, First Edition, November 2025. Editors-in-chief: Zeng Xiaoqin, Zheng Junke, Li Dongliang.)
Translate: Rui Su
Proofread: Mingyuan Sun
