Machine Learning and Protein Engineering Approaches to Understanding Kinesin-5 Activity

Cancer is a term describing a collection of diseases that result in uncontrolled cell growth. Cancer has manifold etiologies and underlying cancers are rouge biochemical pathways involving many different proteins. In the current work, two approaches are used to enhance knowledge of kinesin-5, a potential cancer target involved in cell division. Kinesin-5 promotes cell division by cross-linking and separating microtubules in dividing cells. The first approach uses machine learning (ML) to identify small molecule inhibitors for kinesin-5. Though decades of research have uncovered classes of small-molecules which inhibit kinesin-5 in vitro and in vivo, no candidates have reached phase III clinical trials. In the current work, crystallographic and assay data available for kinesin-5 inhibitors are leveraged to develop ML models for kinesin-5 IC50 predictions. One of the ML models developed is used to screen the Goldilocks subset of the ZINC20 database. Top performing compounds are filtered through a hierarchical clustering approach and ten compounds are presented and their potential binding poses are analyzed. The second approach involves employing a python-based tool, Salt Bridge Builder, to identify putative salt bridges. The highly negatively-charged microtubule proteins, �-tubulin and β-tubulin interact with the positively-charged kinesin-5 to exert their biological function. Salt Bridge Builder is used to determine uncharged-to-charged mutations on kinesin-5 which form salt bridges. Three such mutations which form stable salt bridges in a molecular dynamics (MD) simulation are presented and their effects on the short-distance and long-distance interactions between kinesin-5 and the microtubule proteins are analyzed.