The deep integration of AI and scientific research — AI for Science (AI4S) — has already become an ongoing technological revolution. Compared with traditional scientific methods, machine learning has stronger capabilities in processing high-dimensional data, enabling more effective handling and analysis of large-scale datasets, revealing hidden patterns and laws within the data, and thus providing new opportunities and challenges for scientific research. For example, in the biological field, AI technologies can be used to analyze large-scale genomic data, helping to better understand biological systems and disease mechanisms. In chemistry, machine learning can predict molecular properties and reaction behaviors, offering more efficient methods for the design and development of new drugs. In innovative drug research and development, the first batch of AI-developed drug candidates has entered clinical trials; AI’s ability to propose candidates faster is continuously injecting momentum into innovation. In life sciences, scientists are transforming complex biological problems into computational problems for data analysis, with the prospect of rationally designing optimal solutions — this research strategy, replacing the current trial-and-error approach, exemplifies how AI for Science, especially AI for Life Science, is bringing disruptive impacts to biotechnology and human health.
On February 23, our institution successfully held the first science lecture of the "Next-Post-Future" AI4S monthly series in an online format. Dr. Jin Wengong, a postdoctoral researcher at the Eric and Wendy Schmidt Center of the Broad Institute, was invited to deliver an academic report titled "Towards Unsupervised Drug Discovery Using Geometric Deep Learning." The report was hosted by Professor Tian Yonghong, Dean of the School of Information Engineering. Dr. Jin Wengong is a young scholar with extensive research experience in graph neural networks, graph generative models, geometric deep learning, computational biology, and AI-driven drug discovery. His report showcased the current applications of geometric deep learning in drug research, scientifically envisioned the broad prospects of this field, and provided valuable academic references for researchers in related areas.
In his report, Dr. Jin Wengong introduced several methods to address data sparsity, with a focus on their applications in drug combination prediction and protein-ligand binding prediction. He demonstrated how auxiliary data can be used to improve the accuracy of drug combination prediction models, successfully identifying new drug combinations for treating COVID-19 and pancreatic cancer. Additionally, he discussed the use of unsupervised learning algorithms to build prediction models: by applying SE(3) denoising gradient alignment to the energy model of unlabeled protein-ligand complexes and using the energy function as binding affinity, the effectiveness of protein-ligand binding prediction was verified.
Dr. Jin Wengong’s insightful report attracted over 13,000 professional viewers from related fields, who participated online through live broadcast platforms such as Tencent Meeting, Research Cloud-Bilibili, Research Cloud-Weibo, and Research Cloud-Youdao Dictionary. Online audiences actively interacted with the speaker. This report not only presented Dr. Jin’s latest research achievements in computational biology and AI-driven drug discovery but also provided new research ideas and methods for scholars and researchers in related fields.
The "Next-Post-Future" AI4S series academic activities at Peking University Shenzhen Graduate School, initiated by the institution, mainly consist of two modules: "Exploration Lectures" and "Pengcheng Academic Forums." The former plans to invite one renowned scientist or academic/industrial rising star monthly for exchanges in online or offline formats. The latter will serve as an academic event celebrating the "Anniversary Celebration • Continuing Excellence" of Peking University Shenzhen Graduate School, aiming to gather frontier scientific ideas, foster discoveries in technology and industry, promote the institution’s acceleration in building a multi-disciplinary interdisciplinary research ecosystem driven by AI, and establish new paradigms for AI4S research cooperation.
