HOME > DyRAMO

DyRAMO

Dyramo is a framework for exploring promising candidates (e.g., molecules) by performing multi-objective optimization with a generative model while accounting for the reliability (applicability domain) of multiple property-prediction models, thereby avoiding reward hacking.

Information
Case / Article
Learning Content
Review
Developer’s Voice

Information

Official site	https://github.com/ycu-iil/DyRAMO
Openness	★★★
License	MIT License
Core Developers	Kei Terayama ( Yokohama City University ) Tatsuya Yoshizawa (RIKEN)
Availability	OS: Linux (verified on AlmaLinux 9.3) Python: 3.11 Main dependencies: ChemTSv2 1.0.3 / PHYSBO 2.0.0 Optional dependency: LightGBM 3.2.1 (for property prediction)
Related Papers	Yoshizawa, T., Ishida, S., Sato, T., Ohta, M., Honma, T., Terayama, K. A data-driven generative strategy to avoid reward hacking in multi-objective molecular design. Nature Communications 16, 2409 (2025). DOI: 10.1038/s41467-025-57582-3 Yoshizawa, T., Ishida, S., Sato, T., Ohta, M., Honma, T., Terayama, K. Avoiding Reward Hacking in Multi-Objective Molecular Design: A Data-Driven Generative Strategy with a Reliable Design Framework. ChemRxiv (Version 1, 20 June 2024). DOI: 10.26434/chemrxiv-2024-dh681
Related Sites	GitHub (source code): ycu-iil/DyRAMO Press release (Yokohama City University): YCU Research Portal Press release (RIKEN): RIKEN Reproduction instructions: doc/reproduction_instruction.md