Google DeepMind تضيف مهارات كيمياء لـAntigravity: وكيل ذكاء اصطناعي يقترح تصاميم أدوية جديدة في حوالي ساعة، لكنها فرضيات لازم الباحثين يجربونها في المختبرGoogle DeepMind adds chemistry skills to Antigravity: an AI agent proposes new drug designs in about an hour, but they are hypotheses that need lab testing

On 3 November Google DeepMind announced Cheminformatics Skills for Google Antigravity, aimed at speeding up drug discovery. The company says this kind of workflow usually takes several days per iteration.

In an example using real lung cancer treatments, the agent searched the literature and pulled thousands of compounds within minutes, filtered out irrelevant molecules such as solvents, and documented their sources. It then compared pairs of molecules that differ by a single change, built models to predict potency, and proposed new, more rigid macrocycles with synthesis routes, all in about an hour.

The idea mirrors two well-known drugs for a form of lung cancer: crizotinib has five major rotatable bonds that let the molecule twist into many shapes, while lorlatinib, a third-generation drug that resembles it, is closed into a ring at its end, leaving effectively no rotatable bonds, and is more potent.

The skills are built on the open-source RDKit toolkit and the ChEMBL and PubChem databases, and are available now in Antigravity through the Science plugin. DeepMind cautions that sparse data and lab-to-lab noise limit any model's precision, and that the designs are testable hypotheses, not finished drugs, which must be synthesised and tested in the lab. It is offering 10,000 scientists the top Antigravity tier free for six months, subject to eligibility, capacity and selection.