AI Leaders Gather in Shanghai: Senyi Intelligence, XtalPi, QuanYu Medical, and Honghui Capital Discuss the Future of AI in Healthcare

Since the 1960s, artificial intelligence has experienced three cycles of rise and fall. Currently, AI is undergoing a new wave of explosive growth, with industries across the board actively exploring and developing artificial intelligence.

“Artificial Intelligence + Radiotherapy” enables high-quality medical resources in Beijing, Shanghai, and Guangzhou to be decentralized to grassroots levels; “Artificial Intelligence + New Drug R&D” brings microscopic issues in drug development to light; “Artificial Intelligence + Diagnostic Assistance” not only saves time but also reduces diagnostic error rates.

Since 2015, a new wave of artificial intelligence (AI) has emerged. In the healthcare sector, “AI + medical imaging” was the first to achieve practical implementation. Despite occasional skepticism, it is undeniable that this AI wave is sweeping across the entire healthcare industry. Following “AI + medical imaging,” emerging trends such as “AI + new drug development” and “AI + diagnostic assistance” have successively gained prominence.

At China Renaissance’s annual Healthcare and Life Sciences Leaders Summit, “AI” was undoubtedly the focal point of discussion.

Following AlphaGo’s successive victories over Lee Sedol and Ke Jie, society and the industry began to ponder whether artificial intelligence (AI) would ultimately replace physicians in the medical field. Industry experts from Insilico Medicine, Honghui Capital, Quanyu Medical, XtalPi, and Senyi Intelligence engaged in discussions covering AI’s application scenarios, barriers to adoption, and data ownership in healthcare, seeking to uncover answers to this question.

According to Kang Shigong, Co-founder and Vice President of Quanyu Medical, the two most typical application scenarios for AI in the field of radiation therapy are resource decentralization and quality control.

AI technology has enabled the radiation therapy resources of three hospitals in Beijing, Shanghai, Guangzhou, and Shenzhen to be extended to primary care facilities. “Most patients at the grassroots level lack the financial means to travel to major cities to consult oncology specialists,” he stated. “The integration of AI technology is akin to placing a razor-sharp scalpel into the hands of primary care providers.”

Of course, quality control is particularly critical in this process. By leveraging big data to systematically identify and aggregate all factors contributing to suboptimal quality control, models and algorithms can be developed to automatically monitor and guide the operations of grassroots radiotherapy services.

AI-driven new drug development began to take center stage in 2017, as the integration of artificial intelligence brought microscopic challenges in drug discovery to light.

“It is a tool, akin to a high-resolution microscope.” This is how Li Lipeng, Co-founder of XtalPi and Head of the Big Data and Artificial Intelligence R&D Center, described the application of AI in new drug development. During the process of drug action, it is difficult for humans to visualize how small molecules bind to proteins; however, AI can learn from vast amounts of data to discover the underlying patterns.

In Lai Lipeng’s view, the application of AI in drug development can be described as “building on the past to open up new possibilities.” The vast amount of data accumulated in the past includes data from failed attempts; however, these so-called failures refer specifically to clinical trial failures and do not imply that the data lack value. “Building on the past” means leveraging statistical and machine learning methods to extract previously overlooked insights from historical data.

As for the “pioneering” aspect, deep learning–generated models can help researchers explore a vast chemical space and undertake more groundbreaking work. For instance, a 2017 article in Nature noted that the druggable chemical space may encompass up to 10^60 compounds, whereas currently accessible physical molecular libraries contain only about 10^13 entities.

“There is still a gap of more than 40 orders of magnitude,” said Lai Lipeng. “The drug molecules actually studied in laboratories represent only the tip of the iceberg within the entire chemical space of drugs.” The integration of AI technology enables targeted searches for desired drug molecules within this vast space.

“In addition, I believe AI can also reshape the workflow of new drug development,” he added. “Many existing methods are currently unable to account for the complexity of biological systems.”

For instance, regarding toxicity issues in Phase I clinical trials, the same drug may exhibit markedly different effects in humans and animals. AI emphasizes end-to-end prediction, holding great promise for directly forecasting a series of adverse toxic reactions of candidate drugs in humans based on molecular structure and other factors, thereby significantly reducing the likelihood of failure in clinical drug trials.

“Another critical aspect lies in the development of methodologies for drug discovery and crystal form prediction,” he continued. At XtalPi, they continuously enhance computational efficiency by integrating physics-based models with AI models.

Senyi Intelligence has chosen to focus on the implementation of AI in assisted diagnosis and treatment.

Venous thromboembolism (VTE) is common among patients with prolonged hospital stays, those requiring extended postoperative hospitalization, or those confined to bed for long periods postpartum. This condition carries a risk of progressing to pulmonary embolism, which has a very high mortality rate. To mitigate patient risk, hospitals assign nurses to conduct daily manual monitoring, including disease tracking and feedback for these patients. However, reliance on manual methods consumes significant time and energy and may introduce errors.

“AI technology can improve this situation,” Zhang Shaodian, Founder and CEO of Senyi Intelligence, told VCBeat.

The medical AI products developed by Senyi Intelligence serve two primary functions: first, risk assessment, where the system automatically assigns a risk score to patients based on their clinical condition, medical history (including hospitalization and surgical records), as well as laboratory and diagnostic test results; second, early warning, where patient data is analyzed to predict whether an individual is at high risk. For patients identified as high-risk, alert messages are proactively generated and pushed to physicians’ workstations.

What is the purpose of such products? Zhang Shaodian revealed that, based on over two months of collaboration with tertiary hospitals, these products can help reduce by 95% the time spent on manual patient assessments, while also increasing the identification rate of high-risk patients by 70%.

“This is one of the cases we have attempted in the field of computer-aided diagnosis and treatment,” he added.

Whether in drug development, diagnostic assistance, or radiation therapy, AI technology remains inseparable from big data. Amidst the chaotic and diverse array of data, how can we obtain higher-quality data to achieve better outcomes? Perhaps it is necessary to establish thresholds even before data collection begins.

“The laboratory is the most fundamental setting for data generation, but acquiring data here comes at a high cost,” said Artur Kadurin, Chief AI Officer at Insilico Medicine. Through early efforts, Insilico Medicine has identified methods to acquire foundational experimental data, enabling further expansion of its data resources.

Kadurin believes that vast amounts of data are accessible in China, a feat unattainable overseas. However, Insilico Medicine’s purpose in coming to China extends beyond data acquisition; the company aims to achieve more substantial growth in China and across Asia as a whole.

Li Lipeng agrees with Kadurin’s view that obtaining high-precision, computationally derived data is inherently costly. In its early stages, XtalPi’s drug discovery tool development relied primarily on two sources of data: one was public domain data, and the other was internally generated, high-precision computational data.

The volume of publicly available data is relatively large, but the cleaning process is labor-intensive due to inconsistent data quality and formats. While internal high-precision actuarial data offers high accuracy, its scale can reach hundreds of millions or even billions of records, resulting in significant acquisition costs.

As collaborations with clients deepen, XtalPi has also acquired a portion of data from its partners. These datasets are closely aligned with frontline R&D and specific problem-solving contexts. However, since some of the data were not originally collected for AI modeling purposes, key information may not have been fully documented.

“Undoubtedly, data is the most fundamental element for AI technology,” added He Xing, Partner at Honghui Capital. In addition to being processed and structured, these databases must also enable field extraction and understanding, as well as more intelligent pattern assembly and recognition. At the current stage, such capabilities are difficult to find in China.

“Without the accumulation of such structured data, it is difficult to obtain robust diagnostic results due to the lack of foundational elements,” he continued. He revealed to VCBeat that when evaluating companies, investors place significant emphasis on the sources of corporate data, specifically whether they are legally obtained, whether sufficient de-identification measures have been implemented during use, and whether hospital rights and patient privacy are adequately protected.

“We believe that tech companies should not delude themselves into thinking they own the data.” Zhang Shaodian agreed with this view. He argued that tech companies should leverage their core technologies and capabilities to build formidable competitive advantages, provide better solutions for hospitals, and then utilize a portion of the data within the hospital setting through product adoption.

However, the most critical challenge currently facing AI companies is data governance. Why did artificial intelligence in medical imaging achieve implementation first? Because this type of data is relatively standardized. In contrast, products involving electronic medical records and clinical diagnosis and treatment require extensive data governance. For example, Senyi Intelligence developed a VIE warning and monitoring system for a Grade 3A hospital. For this single system alone, Senyi Intelligence integrated with more than 20 different systems within that hospital.

“Structuring, standardizing, and governing the underlying data is a massive undertaking,” Zhang Shaodian revealed to VCBeat.

As for whether AI will ultimately replace humans, including doctors or nurses within the healthcare system, the answer can already be inferred from the preceding discussion. Currently, AI helps hospitals conserve unnecessary resources and efforts by automating repetitive tasks, reducing turnaround times, increasing output, and lowering misdiagnosis rates. In drug development, AI serves to enhance accuracy and minimize material waste.

Therefore, from any perspective, AI will serve as a method and tool to reduce costs, improve efficiency, and enhance precision across the entire healthcare system. “In the short term, and even for a considerable period thereafter, it is highly unlikely that AI will truly replace humans. However, it can become an excellent assistant,” said He Xing.