Designing MOE Workflows in KNIME for Automated Drug Discovery

FRIDAY, June 29 - Scientific Presentations, 15:00-15:30

George Nicola
Senior Vice President, Computational Pharmacology, Afecta Pharmaceuticals

We have built an automated, workflow-based system that predicts mechanism of action for new indications of safe, off-patent drugs. The platform technology can also design new molecules for a known target or an active drug program. We do this through a combination of enumerating derivatives from a patent, generating a combinatorial library of analogues around a Markush scaffold, chemical fingerprint searches, 3D similarity (shape, pharmacophores, electrostatics), ADMET descriptor matching, gene expression profiling, and protein docking.

The platform is built in the KNIME workflow environment, and uses open source as well as proprietary software, including MOE. The prediction algorithm is custom designed using machine learning models that have been trained on large data sets. We connect and make use of multiple web-accessible databases including those for binding activity, chemical and protein structures, biological pathways, and gene expression.

To feed compounds into the workflow, we have also built a comprehensive compound registration system that analyses, isomerizes, de-duplicates, and uploads compounds to a database server. Our internal library consists of 10,000 commercially available drug compounds, as well as several hundred hand-picked compounds with known activities.

Our workflow-based platform has proven especially useful when partnering with small and mid-size pharmaceutical companies seeking to address an unmet medical need by redesigning an existing product, and where regulatory approval is likely to be achieved rapidly. We provide examples of this platform being used to repurpose molecules into drug candidates.