Flowr.root - A flow matching based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction.

We present Flowr.root, an <i>SE</i>(3)-equivariant flow-matching model for pocket-aware 3D ligand generation with joint binding affinity prediction and confidence estimation. The model supports multiple design modes including <i>de novo</i> generation, interaction/pharmacophore-conditional sampling, fragment elaboration, and multi-endpoint affinity prediction ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow><mml:msub><mml:mtext>pIC</mml:mtext> <mml:mn>50</mml:mn></mml:msub> </mml:mrow> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mtext>p</mml:mtext> <mml:msub><mml:mtext>K</mml:mtext> <mml:mi>i</mml:mi></mml:msub> </mml:mrow> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mtext>p</mml:mtext> <mml:msub><mml:mtext>K</mml:mtext> <mml:mi>d</mml:mi></mml:msub> </mml:mrow> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow><mml:msub><mml:mtext>pEC</mml:mtext> <mml:mn>50</mml:mn></mml:msub> </mml:mrow> </mml:math> ). Training combines large-scale ligand libraries with mixed-fidelity protein-ligand complexes, followed by refinement on curated co-crystal datasets and adaptation to project-specific data through parameter-efficient finetuning. Flowr.root achieves state-of-the-art performance in both unconditional 3D molecule and pocket-conditional ligand generation, producing geometrically realistic, low-strain structures with computational efficiency on established benchmark datasets. The integrated affinity prediction module demonstrates superior accuracy on the Spindr test set and outperforms recent models on the Schrödinger FEP+/OpenFE benchmark while offering substantial speed advantages. As a foundation model, Flowr.root requires continuous parameter-efficient finetuning on project-specific datasets to account for unseen structure-activity landscapes, which we demonstrate yields strong correlation with experimental in-house data. The model's joint generation and affinity prediction capabilities enable inference-time scaling through importance sampling, effectively steering molecular design toward higher-affinity compounds. Case studies validate this approach: selective <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mtext>CK2</mml:mtext> <mml:mi>α</mml:mi></mml:mrow> </mml:math> ligand generation against CLK3 shows significant correlation between predicted and quantum-mechanical binding energies, while scaffold elaboration studies on <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mtext>ER</mml:mtext> <mml:mi>α</mml:mi></mml:mrow> </mml:math> , TYK2 and BACE1 demonstrate strong agreement between predicted affinities and QM calculations. By integrating structure-aware generation, affinity estimation, and property-guided sampling within a unified framework, Flowr.root provides a comprehensive foundation for structure-based drug design spanning hit identification through lead optimization.