Kan AI erstatte 50 % af al lægemiddelforskning ved autonomt at designe og teste nye molekyler in silico ved hjælp af generativ AI og kvantecomputersimuleringer ?

Farmaceutisk forskning og udvikling er notorisk langsom og dyr, men AI accelererer allerede lægemiddelfremstillingen. Hvis en AI ikke blot kunne generere molekyler, men også simulere deres interaktioner med menneskelig biologi i hidtil uset skala, kunne det gøre traditionel laboratoriebaseret forskning forældet. Spørgsmålet er ikke, hvorvidt AI kan designe lægemidler—det er, hvorvidt det kan gøre det bedre end mennesker uden at have brug for menneskelige forskere til at fortolke resultaterne.

Background

Generative AI can today propose novel small-molecule structures with high predicted binding affinity to protein targets, and in-silico high-throughput screening on classical hardware already covers millions of candidates. However, fully autonomous, end-to-end discovery that combines generative design, quantum-grade docking, and lab validation remains out of reach: docking accuracy is still below the ~1 kcal/mol uncertainty needed for reliable affinity ranking, quantum simulations for large proteins are error-prone on near-term devices, and wet-lab synthesis/validation bottlenecks persist. Current demonstrations achieve partial automation (design → in-silico triage → partial synthesis), but no group has reached the 50% throughput reduction threshold across a broad set of targets. SOURCE: McKinsey & Company — https://www.mckinsey.com/capabilities/quantumblack/our-insights/quantum-computing-in-drug-discovery