Insilico Medicine Showcases AI Benchmarking Results on Drug Potency Prediction

A LinkedIn post from Insilico Medicine highlights results from “Day 12” of its ScienceAIBench series, focusing on AI prediction of drug bioactivity. The benchmark centers on IC50, a key potency metric in drug discovery that guides lead optimization by measuring the concentration needed to inhibit a biological target by 50%.

According to the post, the benchmark uses the BindingDB “Cold Drug” split from the Therapeutics Data Commons to test how well large AI models generalize to unseen chemical space. Performance is evaluated using Spearman and Pearson correlations, rather than explanatory R² scores, which reportedly were negative across all models.

The post indicates that the Opus 4.5 model delivered the strongest results, with Spearman and Pearson correlations of 0.347 and 0.349, respectively. GPT 5.2 is described as the next most stable performer, with fewer extreme outliers despite the challenge of predicting activity for novel scaffolds.

Insilico Medicine’s summary notes that the overall predictive signal was moderate but consistent, with correlations peaking around 0.35, underscoring the difficulty of the “Cold Drug” setting. Deepseek 3.2 is reported as the weakest model, with a Spearman correlation of 0.128, suggesting limited generalization to new chemical structures.

For investors, the post suggests Insilico Medicine is positioning itself as a methodical evaluator of cutting-edge AI models for drug design rather than simply a model developer. Demonstrating transparent benchmarking on industry-relevant datasets may strengthen the company’s credibility with pharma partners and could support its ability to monetize platforms or tools that improve lead selection efficiency.

If these benchmarking efforts inform Insilico Medicine’s internal pipelines, even incremental improvements in IC50 prediction could translate into reduced experimental screening costs and faster optimization cycles. Over time, such gains may enhance the company’s competitive stance in AI-driven drug discovery and could be relevant for partnership valuations, platform licensing, and downstream milestone potential.