(October 27, 2016 )
The availability of large, curated, data sets of bioassays from both open-source and commercial vendors, combined with powerful automation frameworks allows unprecedented ability to create predictive models. We used pathway analysis to understand the major pathways for both efficacy and adverse events, then created a large number of QSAR models for activity for proteins in the pathways. The resulting predictive models, and error estimations, were used to screen and rank compounds for both bioactivity and adverse events.
The webinar will discuss using chemistry discovery tools with KNIME for predictive modeling. We will discuss an automated system that creates and assesses large numbers of predictive QSAR models for protein-ligand binding. The example model system is a set of kinase targets including one that may be a toxicity marker for the clinical failure of the JAK2 inhibitor fedratinib.
- How to select the raw bioactivity data to create an evenly distributed data set.
- Use of the Z-test to select statistical significance of active/inactive classification based on model standard deviation.
- Use of KNIME to automate creation and validation of a large set of models.
- Use of KNIME to use the model set to create predicted kinase activity panels for sets of novel compounds.
- Example of using pathway analysis to connect a target with a specific toxicity to create a predictive model for a significant human toxicity.
Matthew Clark, Ph.D.
Consultant, Life Science Services
Matthew Clark, Ph. D. has worked in diverse areas in pharmaceutical research. He has led groups in the scientific software and data industries as well as computational and informatics teams in pharmaceutical companies. His publications span synthetic chemistry, free-energy calculations, informatics and predictive modeling of drug properties, safety pharmacology, and clinical adverse events. As a consultant at Elsevier Dr. Clark works with leading pharmaceutical companies to help address cross functional informatics issues through data integration and predictive modeling. He received his PH.D. in chemistry from the University of Alabama.