Genesis Molecular Ai · 2× · last seen Jul 02, 2026
PEARL (Placing Every Atom in the Right Location) is a generative 3D Foundation Model by Genesis Molecular AI for predicting protein-ligand complex structures. It is based on a diffusion architecture with SO(3)-equivariant neural networks and was trained on large-scale, physics-based synthetic data to overcome the scarcity of experimental structure data. The model was unveiled on October 28, 2025, and according to Genesis, outperforms models such as AlphaFold 3, Boltz-1/2, and Chai-1 in Benchmarks for binding pose accuracy. PEARL is a core component of the GEMS platform (Genesis Exploration of Molecular Space) and is used internally and in pharma collaborations (including with Gilead and Incyte); a public pricing or licensing model for end users has not yet been announced.
Features
| Deployment-Modell | Proprietäres Modell, intern bei Genesis sowie im Rahmen von Pharma-Partnerschaften (z.B. Gilead, Incyte) eingesetzt; kein öffentlicher Self-Service-Zugang dokumentiert |
| Einsatzbereich | Vorhersage von 3D-Strukturen von Protein-Ligand-Komplexen (<1Å RMSD) für Wirkstoffdesign und Arzneimittelforschung |
| Integrationen | Integriert in GEMS-Plattform; NVIDIA cuEquivariance-Kernels für beschleunigtes Training/Inferenz; Kollaboration mit NVIDIA (GTC 2025) |
| Plattform | Teil der GEMS-Plattform (Genesis Exploration of Molecular Space) von Genesis Molecular AI |
| Release-Datum | 28. Oktober 2025 (Ankündigung inkl. technischem Report) |