Insilico Medicine: AI Models Revolutionize Drug Discovery

Insilico Medicine has released new AI models for drug discovery. These models, trained through their MMAI Gym framework, demonstrate state-of-the-art performance on over 50 benchmark tasks in chemistry and biology. What's interesting is that these are language models, but they've been trained with domain-specific precision for complex scientific problems. This marks a shift toward more predictive AI for human health. Historically, specialized computational methods dominated low-data drug discovery tasks. But these new language-model-based specialists are now competing with and even outperforming established approaches. For example, a chemistry model for ADMET prediction achieved top performance across 28 tasks. This includes strong results for drug-drug interaction risk, cytotoxicity, and pharmacokinetic properties. This means it can help identify potential issues earlier in drug development. The bottom line is that these new AI models could significantly speed up and improve the drug discovery process.