HomeArticle

Shortly after receiving the "Young Scientist Award", Xie Weidi secured hundreds of millions of yuan in financing.

36氪的朋友们2026-11-18 10:28
This team with origins in Shanghai Jiao Tong University has made progress in different research tasks in the field of life sciences.

Recently, ChinaVenture has learned that OneX Intelligence, a life model AI company, announced the completion of its first financing round of several hundred million yuan, led by IDG Capital, with CCV, Yunbai Capital, SJTU Hanyuan Asset, and Yunqi Capital participating as follow-on investors, and Palm Capital acting as the exclusive financial advisor. The proceeds from this round will be mainly used for the R&D of foundational life intelligence models, the construction of wet lab platforms, and industrial cooperation validation.

According to data from ChinaVenture CVSource, OneX Intelligence was initiated by Professor Wang Yanfeng from the School of Artificial Intelligence of Shanghai Jiao Tong University, with Zhang Ya and Xie Weidi as co-founders, and Xie Weidi serving as CEO and Chief Scientist. The company's goal is to develop general artificial intelligence for life sciences, and build foundational models that can understand and deduce life processes.

Prior to the official announcement of this round of financing, Xie Weidi was awarded the first "Young Scientist Award" of the World Laureates Association Prize at the award ceremony of the 2026 World Laureates Forum on October 31, becoming one of the six global winners. Earlier, his evidence-based reasoning system for rare diseases DeepRare developed with collaborators was published in *Nature*, and the cell perturbation prediction model MAP was published in *Nature Machine Intelligence*.

This means that from medical evidence reasoning to cell response prediction, this SJTU-origin team has made progress in different research tasks of life sciences. Now, as an entrepreneur, the question OneX Intelligence needs to answer is: can these capabilities be integrated into a unified model system, so that the insights gained from one problem can help solve more problems that have not been fully studied and addressed?

From Computer Vision to Life Sciences

Xie Weidi's research experience and several career shifts are all related to one judgment and cognition: what other problems can AI solve next?

He studied communication engineering for his bachelor's degree, and went to University College London in the UK in 2013 to pursue his master's degree, researching the recognition of handwritten Arabic using computer vision. At that time, the potential of deep learning had not been widely recognized. He taught himself relevant knowledge through open courses, completed the research project, and then entered the Visual Geometry Group (VGG) at the University of Oxford to pursue his doctorate, supervised by Andrew Zisserman and Alison Noble, and won the first Oxford-Google DeepMind Scholarship.

During his research on multimodal self-supervised learning and medical imaging, Xie Weidi gradually shifted his interest to more complex scientific problems. After returning to China and joining Shanghai Jiao Tong University in 2022, the advances in language models and vision models made him start to think: can we leverage these capabilities to further understand the process of disease occurrence and the changes of life systems?

Thus, medicine became Xie Weidi's first stop. "I happen to have a background in medical imaging, so I first entered the healthcare sector, and then gradually dug into the underlying problems of life sciences from the external manifestations of diseases," Xie Weidi said.

This shift is also reflected in the team's support. Co-founder Zhang Ya is currently the Vice Dean and Distinguished Professor of the School of Artificial Intelligence at Shanghai Jiao Tong University. She graduated from the Department of Biological Science and Technology of Tsinghua University for her bachelor's degree, obtained her doctorate from the College of Information Sciences and Technology of the Pennsylvania State University in the US, and used to be a senior R&D manager at Yahoo Labs. Before starting the business, she had long been engaged in interdisciplinary research of artificial intelligence and healthcare, with disciplinary backgrounds in both biology and artificial intelligence, as well as industrial R&D experience.

In this context, as AI reshapes life sciences, the team's research direction has gradually extended from medical problems to life sciences modeling. Among them, DeepRare organizes medical evidence around rare disease diagnosis to form a traceable reasoning process; MAP introduces biomedical knowledge into cell perturbation modeling to explore and predict the response of cells to new interventions. At the same time, the team has also carried out researches on OpenPheno, a model for cell phenotype and biological activity prediction, and KEEP, a knowledge-enhanced pathological foundational model.

According to the team, OpenPheno has been accepted by *Nature Chemical Biology*, and KEEP has been published in *Cancer Cell*.

Behind these achievements is the team's continuous investment in model training, multimodal learning, medical reasoning and life sciences modeling. ChinaVenture has learned that for OneX Intelligence, this is not only the foundation of research, but also connects the company's long-term goals — to link capabilities across different tasks.

What to Do: Foundational Model for Deducing Life Processes

Xie Weidi's judgment on the value of AI is very straightforward: "For things that humans could originally do, AI can do them faster, which improves efficiency; but if it can discover things that humans did not know before, it can push the boundaries of what we can explore outward."

Life sciences provides such research space. The continuous progress of imaging, omics and experimental technologies enables people to observe the local parts of life more and more deeply, but how a single molecular change or drug intervention affects cells and tissues, and finally changes the health state, is still difficult to predict. Disease mechanism research, target discovery, drug design and experimental verification are often carried out in separate stages, making it difficult to connect the knowledge and results of different links.

What OneX Intelligence wants to do is to aim at this direction — "Biological Intelligence". Based on foundational models, the company will connect information at different levels such as molecules, cells and tissues, as well as disease phenotypes and related medical knowledge, so that the model can learn the connections and change rules in living systems.

"The key here is to allow the model to accumulate and transfer capabilities across different research tasks," Xie Weidi said. For example, in the face of an intervention, researchers need to understand which cell functions it will alter, whether it may affect other processes, and what experiments should be conducted next to make judgments. To support such research, the model needs to combine scattered observations, existing knowledge and experimental results, and gradually form the capability to deduce life processes.

On this basis, OneX Intelligence builds a life deduction model with virtual cells as the core through the foundational model, explores and predicts the changes of cell states and functions under different conditions and interventions, and provides support for disease mechanism research, target discovery and drug design.

In Xie Weidi's view, this requires the joint drive of both data and knowledge. "A single experiment records the state under specific conditions. The model also needs to understand the conditions behind the observations and existing mechanisms, and use language models to retrieve literature, integrate knowledge, call tools and organize scientific reasoning, so as to cooperate with capabilities such as cell modeling to assist researchers in proposing hypotheses and planning experiments."

Up to now, in terms of language models, OneX Intelligence has used the current pre-trained models to carry out follow-up training for life sciences, and promoted multimodal integration at the same time. It is reported that relevant R&D will be carried out in coordination with work such as cell modeling and experimental feedback.

In terms of verification, taking rare diseases as the initial scenario as an example, Xie Weidi said that some rare diseases have relatively clear mechanism clues, which are suitable for testing around specific problems whether the model can put forward valuable explanations, predict intervention responses, and provide a basis for target selection and candidate scheme design. This verification work covers mechanism research, target discovery, drug design and intervention effect prediction.

"With the continuous accumulation of cognition and capabilities, the company will apply these capabilities to other diseases and a wider range of life sciences problems."

Integrate Experimental Feedback into Model Iteration

Returning to the most underlying logic, the capability of biological intelligence ultimately needs to be tested by real experiments. In this regard, OneX Intelligence's plan is to build its own wet lab platform, and integrate model R&D, biological observation and experimental verification into the same R&D closed loop.

Xie Weidi said that existing models have integrated existing knowledge, proposed mechanism hypotheses and assisted in designing experiments. Researchers can test the predictions through experiments, correct judgments according to the results, and determine the next research direction. "New experimental results can not only be used to deepen the understanding of mechanisms, but also provide a basis for model training and evaluation, so as to promote the continuous improvement of the model."

In addition, the experimental platform will also carry out observation and measurement around specific scientific problems and model learning needs, so that data generation is closely connected with model R&D. OneX Intelligence hopes to continuously accumulate observation and intervention data corresponding to research conditions through this process, and shorten the cycle from model prediction to experimental feedback.

Such a closed loop also determines the cooperation mode of OneX Intelligence. It is reported that the company will focus on model capabilities and mechanism discovery, and carry out joint R&D with pharmaceutical companies, scientific research institutions, etc. around specific topics, to test whether the model can bring practical value to target discovery, candidate scheme design and experiment selection.

Specifically, DeepRare developed by the team has been deployed in Xinhua Hospital, and is undergoing internal testing, and is planned to be used for rare disease diagnosis and treatment quality control in the future. In the field of drug R&D, OneX Intelligence is also negotiating cooperation with pharmaceutical companies and Contract Development and Manufacturing Organizations (CDMOs) to explore joint research and experimental verification. All these works are at different stages of progress, and the company hopes to gradually establish a transformation path from model capabilities to R&D achievements through specific cooperation.

In terms of business model, OneX Intelligence also hopes to participate in the distribution of achievement value through joint R&D. "If the model discovers targets or candidate schemes worthy of further development, the company will seek to cooperate with partners with subsequent development capabilities to promote the progress." As Xie Weidi said: "For us, what is more attractive is the value brought by new drug discoveries, rather than simply charging based on the volume of model calls."

This pursuit of life sciences exploration also puts higher requirements on the team: model training, scientific reasoning, life system modeling and experimental research need to collaborate around the same problem. With the completion of this round of financing, OneX Intelligence will continue to promote the construction of models and experimental platforms, and also expects outstanding talents in relevant fields to join.

From research breakthroughs in existing tasks, to foundational models that can serve more scientific problems, and then to R&D achievements verified by experiments, what this team aims at is always to establish sustainably accumulative capabilities of biological intelligence in real research, and the foundational model of biological intelligence is universal.

This article is from the WeChat official account "ChinaVenture", Author: Chen Mei, Editor: Wang Qingwu, published with authorization from 36Kr.