FAME 3R: A Fast, Compact, Flexible, and Practical Re-Design of the FAME 3 Model for Predicting Sites of Metabolism

Predicting sites of metabolism (SOMs), i.e., the atom positions in a molecule where metabolic reactions are initiated, is an important component of the computational development pipeline for pharmaceuticals, agrochemicals, and cosmetics. Among the available SOM prediction models, FAME 3, published in 2019, stands out as one of the few noncommercial tools capable of predicting both phase 1 and phase 2 SOMs for a diverse set of xenobiotics. However, significant challenges remained with the original implementation of FAME 3, particularly in terms of maintainability, scalability, and interoperability with other tools, which hindered its broader adoption and long-term usability. To address these shortcomings, we developed FAME 3R, an improved version of FAME 3. Built on Python’s robust data science ecosystem, FAME 3R delivers improved computational efficiency, scalability for larger datasets, and seamless interoperability with contemporary tools. In addition to re-implementing the core functionalities of FAME 3, FAME 3R introduces new features, including support for training models on a broader range of fingerprints and a novel method for assessing prediction reliability using Shannon entropy. FAME 3R is available as an open-source Python package, offering access to all functionalities via an application programming interface (API) and command-line tools. For noncommercial research, pre-trained models can be accessed through the NERDD web server or downloaded for local deployment.