Lian Jye Su, Chief Analyst for Omdia, outlines the key themes of Huawei Analyst Summit 2024 and explores the use of GenAI in the healthcare scenario.
By Lian Jye Su, Chief Analyst, Omdia
Key Takeaways from Huawei Analyst Summit 2024
At HAS 2024, Huawei announced its All Intelligence strategy to over 500 analysts, think tank researchers, and other attendees. The company showcased thought leadership across multiple domains, including foundation models, developer communities, data centers and network equipment, and devices.
As the adoption of intelligence and automation continues to increase, the company’s strategic investment in connectivity and computing has borne fruit, culminating in a strong foundation for the broad scope and application of its All Intelligence strategy. Enterprises that need an optimized AI infrastructure are adopting Huawei’s artificial intelligence (AI) solution, including Pangu models and Huawei Cloud software and services. At the same time, Huawei is helping its telcos and enterprise clients to deploy AI for autonomous operation and maintenance (O&M) for telecommunications and enterprise networks. The company also uses AI internally to boost efficiency.
More importantly, Huawei shared that it proactively engages in global AI governance, ensuring that AI is used for good.
The challenges of modern drug discovery
An area where AI is increasingly important to the well-being of human society is pharmaceuticals, specifically in modern drug discovery. The industry offers solutions to countless life-threatening diseases and pandemics. Unfortunately, traditional drug discovery is very costly. Through manual work and the help of software tools, researchers must identify targets from tens of thousands of molecules, select the right candidate from millions of molecules, and perform rigorous clinical testing to ensure safety and compliance. However, this entire process takes decades and can no longer address the looming healthcare crisis.
As such, the sector is currently suffering from the impact of Eroom’s law, which shows that modern drug discovery has become increasingly slower and more expensive over time despite new technological breakthroughs such as bioinformatics, computational molecular biology, and combinational chemistry.
At the same time, advanced economies are bracing for a silver tsunami, resulting in a loss of economic production and an increasing social welfare burden. On the other hand, new regional pandemics, resistant strains, and antibiotic intolerance are rampaging through emerging and developing countries. Medical staff are running out of options for cost-effective and targeted drugs to rescue communities that are socioeconomically disadvantaged and physically vulnerable. Without effective drug treatment, these economies will be trapped in the vicious cycle of underdevelopment and suffer significantly from opportunity costs. As a result, the gap between the availability and assessability of pharmaceutical knowledge and resources will likely continue to widen when compared to advanced economies.
In-silico drug discovery to the rescue
As such, the industry is actively utilizing AI to overcome this challenge through an approach known as in-silico drug discovery. Combining different data and tools, in-silico drug discovery uses AI models to investigate pharmacological hypotheses and identify new potential drugs.
In-silico drug discovery has been trialed and adopted by many pharmaceutical companies and healthcare institutions worldwide in recent years. Amgen is using AI to accelerate the discovery process and propose and evaluate designs for candidate molecules. Meanwhile, Cleveland Clinic and Francis Crick Institute use it to search genome sequencing findings and large drug-target databases to find effective existing drugs that could help patients with Alzheimer’s and other diseases.
Pangu Drug is Huawei’s attempt to resolve modern drug discovery challenges. Based on Huawei’s Pangu drug molecule model, Pangu Drug uses an encoder-decoder architecture to convert molecular structures into numerical values and learn from these values, allowing the model to predict drug molecule structures and attributes. Some key capabilities include compound-target interaction prediction, compound attribute scoring, molecule generation and optimization, and the reconstruction, validity, and uniqueness of molecular structures.
Currently, Pangu Drug is pre-trained on the chemical structure and attributes of 1.7 billion drug-like molecules. The model can be further finetuned through unsupervised learning or human feedback. The model also features a library of 100 million novel drug-like small molecules that can quickly generate new compounds with similar attributes.
One of the early adopters of Pangu Drug is the No. 1 Affiliated Hospital of the Medical School at Xi’an Jiaotong University. The medical school has managed to design new antimicrobial drugs using Pangu Drug. More importantly, the lead compound development cycle was shortened from several years to one month.
More to be done
For GenAI-based in-silico drug discovery to achieve widespread adoption, the technology must adhere to several shortcomings. First, in-silico drug discovery requires extensive and updated bioinformatics and cheminformatics datasets. Large datasets are essential to the models’ contextual understanding and accuracy of prediction. Second, using AI often results in over-simplification of complex biochemical processes and the flexible nature of protein structure. Third, existing AI models are designed chiefly for optimization and single-target experimentation and perform poorly in multi-target hypothesis testing. However, the industry is hopeful that foundation models that can be pre-trained and finetuned on large domain-specific datasets and are, therefore, much better at contextual understanding, like Pangu Drug, can overcome all the abovementioned limitations.
In addition to AI model advancement, the AI infrastructure designed to host in-silico drug discovery workloads is equally essential. A well-designed, highly optimized system, from AI chips and servers to AI framework, foundation model, software, and applications, can help pharmaceutical companies and healthcare institutions to save energy and reduce their carbon footprint while continuing to discover and develop new life-saving drugs.
Lastly, all AI systems must comply with data security and user safety regulations and requirements. There should not be any compromise on user safety and security for technological advancements. As AI continues to make its way into drug discovery and the broader pharmaceutical industry, Omdia expects healthcare regulators, pharmaceutical companies, public and private healthcare organizations, health practitioners, and industry vendors, like Huawei, to work together to share best practices and define frameworks on the utilization of AI in pharmaceutical.
In summary, Omdia believes that Pangu Drug is another excellent example of AI for good, and represents a strong use case within the broad vision of Huawei’s All Intelligent strategy. AI should always be used for human society’s betterment and well-being. Innovative GenAI-based solutions like in-silico drug discovery will help significantly bridge the socio-economic gaps and usher in a future with more inclusivity, equality, and fairness.
Appendix
Further Reading
2024 Trends to Watch: AI in Healthcare (January 2024)
Artificial Intelligence Software Market Forecasts – 2H23 Data (November 2023)
Healthcare AI Maturity and AI Budget Survey Analysis – 2023 (June 2023)
Quantum Computing in the Life Sciences – 2023 Omdia (September 2023)
Author
Lian Jye Su, Chief Analyst, Applied Intelligence
CONTACT US
omdia.com
askananalyst@omdia.com
