Recently, the concept of physics-based AI has swiftly garnered significant attention within the realms of AI4S (AI for Science) and pharmaceuticals. A multitude of startups in this domain have embraced it as their central focus, highlighting its importance in their funding announcements. Even industry behemoths like NVIDIA and Eli Lilly are actively engaging in strategic maneuvers, with over US$3 billion already invested in this burgeoning sector.

At present, a universally accepted definition of physics-based AI within the pharmaceutical industry remains elusive. However, those venturing into the life sciences arena can be broadly classified into two categories. The first type operates within physical environments, concentrating on laboratory intelligence and harnessing intelligent robots and other equipment to augment experimental efficiency. The second type integrates physical knowledge into models, employing physical laws such as molecular dynamics and quantum chemistry to construct models, thereby diminishing the reliance on actual experiments through preliminary simulations.

The burgeoning interest in physics-based AI stems from the practical challenges currently confronting AI-driven drug discovery. Existing AI models predominantly rely on statistical patterns derived from historical samples, performing inadequately in critical areas such as molecular affinity prediction. Furthermore, the industry generally lacks comprehensive experimental data that includes both successful and unsuccessful cases, making it arduous to establish an effective research and development closed loop (a term denoting a self-contained system where feedback from outcomes influences subsequent processes). By gaining access to real experimental scenarios and data, physics-based AI4S presents itself as a promising avenue for surmounting the development bottlenecks in AI-driven drug discovery.