PaccMannRL: De novo generation of hit-like anticancer molecules from transcriptomic data via reinforcement learning

• DOI: 10.1016/j.isci.2021.102269
• Publication Date: 2021-03-07T20:15:03Z
• Authors: Jannis Born; Matteo Manica; Ali Oskooei; Joris Cadow; Greta Markert; María Rodríguez M...
• Keywords: Cancer cell lines

With the advent of deep generative models in computational chemistry, in-silico drug design is undergoing an unprecedented transformation. Although learning approaches have shown potential generating compounds with desired chemical properties, they disregard cellular environment target diseases. Bridging systems biology and design, we present a reinforcement method for de novo molecular from gene expression profiles. We construct hybrid Variational Autoencoder that tailors molecules to target-specific transcriptomic profiles, using anticancer sensitivity prediction model (PaccMann) as reward function. Without incorporating information about drugs, molecule generation biased toward high predicted efficacy against cell lines or cancer types. The can be further refined by subsidiary constraints such toxicity. Our cancer-type-specific candidate drugs are similar drug-likeness, synthesizability, solubility frequently exhibit highest structural similarity known these