In Silico ADME/Tox in the Generative AI Paradigm
Sean Ekins, Thomas R. Lane, Joshua S. Harris, Fabio Urbina · 2024
Computational approaches for absorption, distribution, metabolism, excretion, and toxicity (ADME/Tox, also known as ADMET) prediction have been developed over several decades. These were initially used for predictions of such properties to reduce late-stage clinical failures. We have progressed from optimizing the desired bioactivity and an ADME/Tox property to multiple optimization which combines many factors simultaneously. As we now embark on the generative de novo design paradigm, many ADME/Tox properties can be optimized in parallel with bioactivity to produce molecules with desired characteristics. Clearly, some of these ADME/Tox models may be better (and bigger) than others, and this will also create challenges that will be discussed.