PaccMannRL: Designing anticancer drugs from transcriptomic data via reinforcement learning
Jannis Born, Matteo Manica, Ali Oskooei, Joris Cadow, María Rodríguez Martínez
Received Date: 6th April 20
With the advent of deep generative models in computational chemistry, in silico drug design has undergone an unprecedented transformation. While state-of-the-art deep learning approaches have shown potential in generating compounds with desired chemical properties, they disregard the cellular environment and biomolecular properties of the target disease. Here, we introduce a novel framework for de-novo molecular design that systematically leverages systems biology information into the drug discovery process. Embodied through two separate Variational Autoencoders (VAE), the drug generation is driven through a disease context (transcriptomic proﬁles of cancer cells) deemed to represent the target environment in which the drug has to act. Showcased at the challenging task of de-novo anticancer drug discovery, our conditional generative model is demonstrated to be capable of tailoring anticancer compounds to target a speciﬁc biomolecular proﬁle, according to the critic. Without incorporating explicit information about anticancer drugs, we demonstrate how the molecule generation, starting from a random point in a chemical space, can be biased towards compounds with high predicted inhibitory eﬀect against individual cell lines or cell lines from speciﬁc cancer sites. We verify our approach by investigating candidate drugs generated against speciﬁc cancer types and ﬁnd the highest structural similarity to existing compounds with known eﬃcacy against these cancer types. Despite no direct optimization of other pharmacological properties, we report good agreement with known cancer drugs in metrics like drug-likeness, synthesizability and solubility. We envision our approach to be a step towards increasing success rates in lead compound discovery and ﬁnding more targeted medicines by leveraging the cellular environment of the disease.
Read in full at arXiv.
This is an abstract of a preprint hosted on an independent third party site. It has not been peer reviewed but is currently under consideration at Nature Communications.