As artificial intelligence (AI) technology becomes increasingly widespread in the pharmaceutical industry—spanning drug R&D, production optimization, market strategy formulation, and pharmacovigilance—AI is gradually shaping the future of the pharmaceutical sector. Drawing on key insights from PharmaRong Consulting’s *2023 White Paper on AI-Driven Pharmaceutical Companies in China*, this article explores how AI is driving the optimization of drug production processes, supporting innovation in pharmaceutical market strategies, and advancing applications in pharmacovigilance, while also examining the strategic initiatives and progress of domestic companies in these areas.

I. AI-Empowered Drug Production Support

1. Examples of the Latest Technologies

Artificial intelligence also offers numerous possibilities for drug manufacturing, including but not limited to optimizing process design and control, intelligent monitoring and maintenance, and trend monitoring to drive continuous improvement. Deploying AI to support drug manufacturing can be integrated with other advanced production technologies to achieve the desired benefits. AI serves as a catalyst for implementing Industry 4.0, enabling manufacturers to create a well-controlled, hyper-connected, and digital ecosystem across the pharmaceutical value chain.

According to the discussion paper “Artificial Intelligence in Drug Manufacturing” released by the FDA’s Center for Drug Evaluation and Research, the application of AI in drug manufacturing can be broadly categorized into four scenarios: process design optimization and scale-up; advanced process control; process monitoring and defect detection; and trend analysis and detection.

Currently, the application of AI in drug manufacturing is still in its early stages, but some interim achievements have been made. For example, the biotechnology company Pow.bio is moving toward optimizing and automating fermentation through an AI-powered continuous fermentation platform, making the process both cost-effective and streamlined.

Teledyne DALSA has developed a specialized Braille visual inspection system. This system utilizes the shape-shadowing technology of the VICORE smart vision system—equipped with a line-scan camera—along with the Sherlock shape-shadowing algorithm to capture high-contrast 3D images from Braille text with complex backgrounds. Subsequently, Sherlock applies preprocessing to optimize the shape of missing Braille dots, enabling OCR algorithms to read Braille characters and assist the visually impaired in reading medication instructions.

Digital twin technology can be adopted in process optimization design. A process digital twin is a digital replica of a physical process used to better understand, analyze, predict, and optimize process performance. Digital twins are particularly useful for analyzing manufacturing processes characterized by limited development data. For example, GSK collaborated with Siemens [1] to successfully validate digital twin technology at the pilot scale;The Austrian startup Novasign developed a hybrid-model-based digital twin system, which was used to optimize the process of E. coli expressing superoxide dismutase [2], accelerating the fermentation process optimization; In 2019, Siemens acquired PSE’s process digital modeling software platform gPROMs to establish a digital twin system for bioprocesses.

Key Architecture of Digital Twins

In pharmaceutical manufacturing, AI/ML technologies such as neural networks can be used to implement APC (Adaptive Process Control) using real-time process data as input. These methods can also be applied to develop process controls capable of predicting whether a process is operating within a controlled state. This requires combining real-time sensor data with AI/ML tools, including integration with intelligent production line monitoring, to improve the efficiency and yield of existing manufacturing lines.

For pharmaceutical quality control, machine vision inspection technology is frequently employed. Inspecting products with inherent challenges during pharmaceutical production is a highly demanding task. For example, distinguishing bubbles from particles in high-viscosity injectables where bubbles cannot be completely removed is difficult. Achieving a balanced detection level and false rejection rate typically requires extensive development and optimization of vision algorithms."AI image vision inspection" essentially involves using deep learning-based AI algorithms to capture images of the inspection objects, followed by sample annotation and training, to achieve the goal of classification decision-making—separating good products from defective ones.

For trend monitoring, AI/ML can be used to assist in reviewing deviation reports, which typically contain large amounts of data or text. This enables the analysis of manufacturing-related deviation trends, the clustering of problem areas, and the prioritization of areas requiring proactive, continuous improvement.AI/ML methods integrated with Process Performance (Ppk) and Process Capability (Cpk) metrics can be used to proactively monitor trends and out-of-control events in manufacturing operations, and to predict thresholds that trigger CAPA (Corrective and Preventive Action) effectiveness assessments.

2. Domestic Industry Landscape

According to statistics from PharmaRong Consulting, few domestic companies currently apply AI technology to pharmaceutical manufacturing processes. Key "AI+Pharma" companies include Woshi Technology, Dawan Bio, and Shengpu Zetai. While these three companies operate in different research fields, their AI applications in drug manufacturing are primarily focused on process optimization.

Woshi Technology focuses on small-molecule compounds, concentrating its efforts on AI-driven retrosynthesis and process optimization. Its primary clients are CROs and CDMOs, assisting these companies in synthesizing molecules and optimizing their synthesis processes.

Dawan Bio primarily applies AI technology to the biopharmaceutical sector. The company has three intelligent bioprocess development platforms that have entered commercial operation: Klone4.0™, AlfaStaX®, and AlfaMedX®. By integrating AI with biotechnology, Dawan Bio provides CDMO services for antibody and protein-based drugs, developing site-specific integration high-expression cell lines through the combination of AI and bioprocessing.

Shengpu Zetai is dedicated to the R&D and production services for peptide drugs. The company has developed its big data and AI-driven "Chemical Space" drug discovery technology in Switzerland for nearly 20 years and has collaborated with numerous pharmaceutical companies, including Roche, Lonza, Bracco, Renfu Pharmaceutical, and Xingqi Eye Pharmaceutical.

China’s pharmaceutical industry boasts high annual output and a wide variety of products. During the production and aluminum-plastic packaging of capsules and tablets, defects such as missing granules, dents, breakage, and illegible batch numbers are inevitable. In the pharmaceutical manufacturing process, in addition to the need for process optimization, there is an urgent demand for quality control of pharmaceutical products.Some companies that leverage artificial intelligence to enhance productivity operate across a broad range of sectors, including the life sciences. Examples include Magia Technology and Keyi Technology; such companies primarily provide defect detection services for pharmaceutical products.

Additionally, the Taicang Institute of Information Technology at the Chinese Academy of Sciences has partnered with Huawei to launch the Zhimiao series of intelligent vision inspection products. Based on the Ascend AI hardware and software platform, these products integrate AI model algorithms such as transfer learning, data augmentation, and weakly supervised detection. They address the challenge of universal inspection for blister-packaged pharmaceuticals across different products, models, and categories, delivering a digital and intelligent solution for pharmaceutical packaging defect detection.

The core products of the ZhiMo series of intelligent vision inspection systems include:

A. ZhiMo ZMAI Visual Inspection Software: Based on the MindSpore AI framework and MindX deep learning vision SDK;

B. ZhiMo Industrial AI Visual Inspection Platform: Enables rapid training and iteration on large-scale data to build a comprehensive defect detection model library;

C. ZhiMo Pharmaceutical Inspection Equipment and Modules: Based on the Ascend Atlas 300 AI inference card, the standalone ZMX-100 inspection device performs comprehensive detection of packaging defects on both sides of blister packs as well as drug defects; the standalone ZMX-320 vision inspection module can be mounted on blister packaging machines to achieve effective improvements in recognition accuracy while maintaining high-speed, high-throughput inspection.

MindSpozre Overall Architecture

II. AI-Empowered Market Expansion and Commercialization of Pharmaceuticals

1. Examples of the Latest Technologies

AI marketing refers to the process of utilizing artificial intelligence technologies and algorithms to assist and improve marketing activities. It combines technologies such as big data analysis, machine learning, and natural language processing to provide personalized, precise, and intelligent marketing solutions through the processing and analysis of massive amounts of data.In AI marketing, artificial intelligence can help enterprises achieve objectives such as market analysis and forecasting, target audience segmentation, personalized marketing and promotion, marketing automation, customer relationship management, and marketing decision support. These applications enable pharmaceutical companies to better understand market and consumer needs, thereby delivering personalized marketing and promotional strategies.

Furthermore, a key role of AI in pharmaceutical market development is to assist with product pricing. The pricing methodology leverages AI’s ability to mimic human expert thinking to evaluate factors influencing post-production pricing. Factors determining the prices of innovative and generic drugs include: expenditures during the drug R&D process, price regulatory systems in relevant countries, the duration of exclusivity periods, the market share of innovative drugs one year after patent expiration, the prices of reference products, and pricing policies.

Currently, companies such as Intelligence Node, Veeva Systems, Aktana, and DeepIntent have already applied AI to pharmaceutical market development and commercialization. For example, Intelligence Node’s “In-Competitor” is a retail competitive intelligence platform that analyzes competitors’ pricing data to help retailers and brand owners monitor competitors.

The AI technologies used in AI-driven pharmaceutical market development primarily include NLP, ML, data mining, and big data analysis. In ML, software analyzes vast amounts of statistical data—such as product development costs, market demand, inventory costs, manufacturing costs, and competitors’ product prices—to develop algorithms that predict product prices.

2. Domestic Industry Landscape

In 2023, heightened global economic uncertainty, market pressures, and intensified competition have made the issues inherent in traditional pharmaceutical marketing models—such as high costs, low efficiency, poor conversion rates, and limited methods—even more apparent. This has compelled companies to accelerate the transformation of their marketing models and technological strategies to establish a second growth curve.Artificial intelligence technology enables a more accurate understanding of patient needs, thereby helping to determine more appropriate pharmaceutical marketing strategies. AI-based analysis can also help pharmaceutical companies shorten market feedback cycles, thereby optimizing drug promotion strategies.

Against this backdrop, China has made significant progress in AI-enabled pharmaceutical commercialization. According to statistics from Yaorong Data, there are currently six companies in China dedicated to empowering market development and commercialization through AI. By categorizing their service offerings and application scenarios, we find that these companies’ services can be broadly divided into: communication with Key Opinion Leaders (KOLs), market access strategy formulation, sales forecasting, competitive analysis, and compliance with sales messaging regulations.

III. Gradual Application of AI in Pharmacovigilance

Pharmacovigilance (PV): Pharmacovigilance refers to the scientific research and activities related to the detection, evaluation, understanding, and prevention of adverse drug reactions or any other potential drug-related issues.

Currently, AI plays a very important role in PV management.AI, including ML methods such as natural language processing and deep learning, can detect and extract information on adverse drug events, thereby automating the pharmacovigilance process and improving the monitoring of known and documented adverse drug events. Furthermore, with the increasing demand for telemedicine services, AI can play a role in detecting and preventing adverse drug events in the management of acute and chronic diseases. Strengthening the drug safety regulatory system with the support of data intelligence technology is a crucial component of improving healthcare security.

Harvard University conducted a scoping review of the use of ML-based AI, searching PubMed, Embase, Web of Science, and IEEE Xplore databases to identify articles published between 2000 and September 2021 related to the use of ML in pharmacovigilance. The results showed that the majority of studies (53%) focused on using traditional statistical methods to detect safety signals.Among studies employing newer ML methods, 61% utilized off-the-shelf technologies with minor modifications. A temporal analysis revealed that newer methods, such as deep learning, have been increasingly adopted in recent years. Advances in AI have not yet fully permeated the pharmacovigilance field (although recent research suggests this may be changing). Opportunities for implementing ML methods exist throughout the entire pharmacovigilance pipeline.

The FDA began applying AI to pharmacovigilance (PV) relatively early to enhance the efficiency and scientific value of Investigational Drug Safety Reports (IVSRs) analysis. In addition to the nearly 2 million FAERS reports received annually from the pharmaceutical industry, the FDA processes hundreds of thousands of reports submitted directly by the public and transfers them to the FAERS database.As described in the account of the FDA’s progress, most AI applications in pharmacovigilance involve the use of NLP (natural language processing) to automatically extract key features relevant to causal association assessments from ICSR narratives. A smaller number of efforts aim to develop predictive machine learning (ML) algorithms to automate the human cognitive processes involved in extracting, integrating, and analyzing key information elements from ICSRs.

Table 1 summarizes the major advancements in the FDA’s application of artificial intelligence to pharmacovigilance activities in recent years:

1. Domestic Industry Landscape

According to the published 2022 National Adverse Drug Reaction Monitoring Annual Report, the National Adverse Drug Reaction Monitoring Network received 2.023 million adverse reaction reports in 2022; from 1999 to 2022, the network accumulated a total of 20.856 million adverse reaction reports [1].Given the massive volume of ICSRs (Individual Case Safety Reports) received annually, how to process, analyze, evaluate, and utilize this vast and heterogeneous dataset, and how to better leverage ICSRs to support post-marketing drug regulatory work, represent one of the major challenges facing China’s pharmacovigilance (PV) efforts. The two main components of a PV system include individual case processing—that is, the collection, interpretation, and reporting of individual adverse events—and signal detection, which involves continuous trend monitoring of AE data to determine whether any unknown safety information may impact the results of risk-benefit assessments.

China began applying artificial intelligence to PV later than Western countries, but currently, pharmacovigilance systems based on real-world data and data intelligence technologies are reshaping the landscape of drug regulation in China.

Currently, there are approximately seven companies in China providing pharmacovigilance services using AI technology; most of these are concentrated in Beijing and operate as SaaS providers.Apart from Dongshi Network, other companies offer services beyond pharmacovigilance that cover other stages of the drug development lifecycle. Founded in 2016, Dongshi Network specializes in pharmaceutical safety and adopts a model of “SaaS-based digitalization of the pharmaceutical safety industry across the entire lifecycle and supply chain, combined with highly specialized vertical safety services.” It serves the pharmaceutical industry by providing pharmacovigilance and risk management services covering the entire lifecycle from clinical trials to post-marketing.

List of Domestic AI+Pharmacovigilance Companies

Conclusion:

The deep integration of AI technology into the pharmaceutical industry has not only driven dual improvements in production efficiency and quality but has also revolutionized market expansion strategies while enhancing the scientific rigor and response speed of pharmacovigilance. The active exploration by domestic companies in these areas not only demonstrates the immense potential of AI technology but also offers new insights for the global pharmaceutical industry. In the future, with continuous technological advancements and deeper application, AI will continue to propel the pharmaceutical industry toward greater intelligence, efficiency, and safety.