Conference paper Open Access

PaccMannRL: Designing anticancer drugs from transcriptomic data via reinforcement learning

Born, Jannis; Manica, Matteo; Oskooei, Ali; Cadow, Joris; Rodríguez Martínez, María

With the advent of deep generative models in computational chemistry, in silico anticancer 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 genetic profile and properties of the target disease. Here, we introduce the first generative model capable of tailoring anticancer compounds for a specific biomolecular profile. Using a RL framework, the transcriptomic profiles of cancer cells are used as a context for the generation of candidate molecules which is optimized through PaccMann (a previously developed drug sensitivity prediction model) to obtain effective anti-cancer compounds for the given context (i.e., transcriptomic profile). 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. We envision our approach to transform in silico anticancer drug design by increasing success rates in lead compound discovery by leveraging the biomolecular characteristics of the disease.

Files (3.2 MB)
Name Size
paccmann_rl_neurips_mlpw.pdf
md5:2f4c29e6dd81eea3f799e4f4d509fa72
3.2 MB Download
26
7
views
downloads
All versions This version
Views 2626
Downloads 77
Data volume 22.5 MB22.5 MB
Unique views 2424
Unique downloads 66

Share

Cite as