Purpose

The growing imperative to minimize animal testing has driven the adoption of New approach methodologies (NAMs), including read-across (RAx) methodologies. Identifying truly equivalent reference compounds for TK read-across is challenging, as structural similarity alone does not ensure toxicokinetic equivalence. We developed a novel framework for TK RAx that doesn’t use new animal studies. By assuming TK equivalence is defined by plasma concentration-time (Cp-time) curve similarity, our method uses a Recurrent Neural Network (RNN) autoencoder to compress predicted high-throughput-pharmacokinetic curves (HTPK) into a latent space, finding nearest neighbors for comparison with traditional chemical descriptors and TK inputs.

By Priyata Kalra, Rafał A. Bachorz, Michael Lawless

Eurotox 2025, September 14-17, 2025, Athens, Greece