DeepSeek is penetrating into various industries. In February 2025, the domestic large model DeepSeek-R1 was fully open source and adapted to multiple fields, and AI medical care also entered a new stage of technology integration and industry reconstruction. Many companies including Aimi Vaccine, Hengrui Medicine, and Zhiyun Health announced that they are combining with AI large models such as DeepSeek.

Take the recent announcement of Aimi Vaccine as an example. It stated that as a leading vaccine company in China, the group actively responded to the national "AI+" industrial development strategy and fully deployed access to the DeepSeek large model. However, the financial report of Aimi Vaccine shows that it has faced continuous losses in recent years. To what extent can DeepSeek reduce costs and increase efficiency?

Many industry analysts consulted by Blue Whale Finance all said that overall, they are optimistic about the application of AI big models such as DeepSeek in various application scenarios in the future, but there are still challenges such as insufficient accuracy, error correction costs may be greater than speed-up costs, and the industry needs vertical models more.

DeepSeek still has limitations in accelerating pharmaceutical research and development, and the industry is more inclined to vertical models

In a recent announcement, Aimi Vaccine pointed out that in order to build an intelligent vaccine research and development system, the group will use the DeepSeek big model to obtain better candidate vaccines through discovery, design, manufacturing and other processes. Integrating multi-source heterogeneous data from different bioinformatics databases, by fusing pre-trained protein language models with graph neural network architectures, it is expected to achieve intelligent screening of antigen targets and dynamic analysis of conserved regions. At the same time, by accessing real-time mutation monitoring data streams, using long and short-term memory networks to predict the future mutation direction of pathogens, dynamically analyzing conserved regions and predicting broad-spectrum antigen epitopes. The group will also use DeepSeek to combine generative models with deep learning technology to optimize mRNA vaccine sequence design and stability. By predicting the secondary structure of mRNA, according to the codon preference characteristics of gene expression in host cells, efficient component combinations are intelligently designed to improve the stability and translation efficiency of mRNA and generate the optimal sequence. In addition, the group is exploring the use of DeepSeek's AI framework to optimize process parameters to improve antigen production and quality.

In short, according to the above announcement, DeepSeek seems to be changing all stages of drug development.

Aime Vaccine is known as one of the "Three Musketeers of Domestic mRNA Vaccines" because it completed the acquisition of Zhuhai Livanda Biotechnology Co., Ltd. at the end of May 2021. However, although the mRNA track quickly broke the circle during the COVID-19 pandemic, with the end of the COVID-19 pandemic, the mRNA track is also facing a rapid decline.

In terms of performance, Aime Vaccine has been facing continuous losses in recent years.

Its financial report shows that from 2021 to 2022 and 2023, Aime Vaccine's revenue was 1.57 billion yuan, 1.264 billion yuan, and 1.187 billion yuan, respectively, and its net losses attributable to the parent company were 692.8 million yuan, 319.6 million yuan, and 1.301 billion yuan, respectively. Although Aime Vaccine has not yet announced its full-year performance for 2024, from the semi-annual report, although the loss has narrowed, its net loss attributable to the parent company is still negative, at -139.3 million.

The reason for its significant loss in 2023 is mainly due to the changes in the global COVID-19 pandemic, and the mRNA COVID-19 vaccine pipeline and technology platform acquired by Livanda Bio have suffered a large amount of impairment, with an estimated impairment loss of between 1.65 billion yuan and 1.8 billion yuan. In addition, Aime Vaccine's R&D expenses in 2023 are 636 million yuan.

In fact, R&D expenses have always been one of the main expenses in the medical industry. So to what extent can access to DeepSeek reduce this part of the expenses? Can its R&D pipeline show explosive growth? Blue Whale Finance sent an email to Aimi Vaccine for an interview, but no response was received as of press time.

Not only Aimi Vaccine, but also many pharmaceutical companies including Hengrui Medicine, Yunnan Baiyao, and Weining Health have recently announced access to DeepSeek.

Zhou Mingzi, Executive Director of Sullivan Greater China, told Blue Whale Finance that the application of large models such as DeepSeek has brought new investment opportunities to the AI medical field in China and even the world, and has significantly brought new attraction and investment confidence to the capital market. Sub-sectors, such as medical image analysis, personalized treatment plans, telemedicine, intelligent auxiliary diagnosis and drug development, will become hot spots for AI medical investment. Innovative applications in these fields can not only attract investment enthusiasm from the capital market, facilitate financing for innovative companies, accelerate the innovation and commercialization of AI products in the medical field, but also further accelerate the overseas expansion of China's local AI medical innovation business. Against this background, investment opportunities around products, businesses, and services overseas will also grow rapidly in the future.

However, Sha Lin, IT director of Insilicon Intelligence, who has been deeply involved in the field of AI pharmaceuticals for many years, told Blue Whale Finance that Deepseek is indeed a groundbreaking product. On the one hand, it has achieved technological breakthroughs, breaking the monopoly of the GPT 01 model, and published detailed papers to let the industry fully understand the training path of the model; on the other hand, it has achieved technological equality. It adopts the most relaxed open source protocol, and even some areas where LLM development is relatively backward can obtain the most advanced technology through open source code.

"Deepseek is a general model like GPT, and more proprietary vertical models are used in the field of drug research and development. Simply from the data point of view, the general model has a larger amount of data, and the vertical model has a more precise amount of data, so in terms of the output results, the former can solve some more general problems, while the latter, in addition to using proprietary data sets, also has professional team feedback during the model training process, so that the output results are more credible and professional. Therefore, we believe that Deepseek's direct help to drug research and development may not be significant." Sha Lin said.

In addition, some industry insiders told Blue Whale Finance that, in fact, the field of pharmaceutical research and development has long been connected to large models. However, the logic of proprietary model training requires expert feedback in addition to professional data. Deepseek is very easy to correct errors if it answers a broad question incorrectly, and your feedback is the process of its training; but for example, if you ask what sites can bind to the TNIK target, Deepseek will list a bunch of them for you, and it is difficult for you to correct them. There is a cost for error correction. You can see it yourself with cryo-electron microscopy, etc. The cost of expert feedback here is very high, so it is not suitable for such training for Deepseek.

"Overall, the accuracy and error correction cost of drug development and Deepseek access are a major challenge." said the above person.

Breakthroughs in multiple scenarios in the future may reshape the medical and health industry

Although there are still challenges, many industry consultants told Blue Whale Finance that the breakthrough progress of artificial intelligence technology is reshaping the medical and health industry in a subversive manner.

In November 2024, the "Reference Guidelines for Artificial Intelligence Application Scenarios in the Health Industry" was released, which clarified 84 application directions, covering core areas such as medical image analysis and intelligent drug development, providing top-level design support for the implementation of technology.

As for what are the application scenarios of AI in the medical industry in the future?

According to the research report of Du Xiangyang's research team at Southwest Securities, the ecological closed loop of "model as a service" not only reconstructs the medical workflow, but also spawns the outbreak of six core application directions, including AI medical image-assisted diagnosis; AI gene sequencing, that is, rare disease large model optimizes gene sequencing screening, AI assists multi-study data analysis; AI medical information CDSS assists clinical decision-making, that is, optimizes electronic medical record quality control, clinical decision support and intelligent triage process; AI health management, that is, the platform relies on the model to generate personalized prevention and control plans, and promotes the management of chronic diseases from "passive treatment" to "active intervention"; AI pharmaceutical manufacturing, that is, shortening the drug development cycle, assisting target discovery and clinical efficacy prediction; AI surgical robots.

Chen Chen, head of China's medical industry research at UBS Securities, told Blue Whale Finance that we have noticed that AI-driven drug discovery has always been a hot topic. Different AI tools have been widely used in the biopharmaceutical and CRO industries to enable drug-target interactions, inhibition mechanisms, metabolic simulations, etc. Some pharmaceutical and CRO companies also apply DeepSeek and other AI big models to their daily operations and clinical trial management processes such as data management. Some companies believe that DeepSeek can help improve operational and R&D efficiency. Auxiliary diagnosis (such as imaging analysis and enhancement) is adopting AI tools to help improve diagnostic quality.

"We have noticed that surgical robots are also using artificial intelligence to assist surgeons in enhancing their surgical vision and surgical experience. Internet medical companies use AI technology to improve the efficiency of doctors during medical consultations and the quality of online health management services; AI improves productivity and customer satisfaction while controlling risks and human errors. In retail pharmacies and medical distribution, artificial intelligence can be used to improve logistics efficiency, inventory turnover, operational management, and customer service quality. Some healthcare AI applications focus on clinical decision support and other aspects, including: intelligent decision support: auxiliary diagnosis and auxiliary screening, intelligent alarm, and comprehensive analysis of patient status; clinical automation: automatic shift handover, medical record writing, and voice-to-text dialogue; telemedicine: unified standards and automatic quality control; intelligent management: quality control, operational performance, and scientific research." Chen Chen said.

Dong Xiaoya, executive director of CIC, said that domestic large models have made a lot of gratifying progress from catching up with overseas to now. The performance and cost of large models have reached the world's leading level, and have the foundation for large-scale commercial use. Driven by this, the scale of China's AI large model industry has also increased rapidly. At present, the disadvantages of China's large model industry are mainly reflected in hardware infrastructure and basic algorithm theory. China's high-performance chips are still behind the international leading level, and the innovative theoretical research of basic algorithms is relatively backward. However, the success of DeepSeek has also provided confidence for industry participants. In the future, through the joint efforts of all sectors, breakthroughs will be made in these aspects.

https://finance.eastmoney.com/a/202502253329303697.html