Drug Candidate Screening and Effective Descriptors for SVMs
My coworkers and I devised a method to effectively select drug candidates at the Drug Discovery Lab1 of CrystalGenomics, Inc. under Dr. Seonggu Ro (as a full-time intern: December 2003–February 2004; as a part time-intern: September 2002–November 2003). I researched on potential drug descriptors that might determine the suitability of a chemical as a drug in conjunction with the Support Vector Machines (SVMs). I also implemented an automated system that specifically selects most suitable drug candidates out of a huge chemical database with other Ph.D. researchers (used languages: C++, Python, SQL). In addition to the drug discovery research, I visualized the 3D conformations of the protein-substrate binding complexes, especially around active sites, using VMD and Insight II. Here are some pictures:
Figure 1: A thrombin molecule with a beta-strand mimetic inhibitor (yellow balls). Red and blue balls denote positively and negatively charged side chains respectively. The transparent surface represents the surface of the inhibitor and the sign of the electrostatic potential on it (red: positive; blue: negative).
Figure 2: A thrombin molecule with the inhibitor (different view). (a) The brown and green surface denotes the surface of the inhibitor with the sign of the potential (brown: positive; green: negative). (b) The blue surface represents the surface of the thrombin molecule. The inhibitor fits well in the pocket.
Notes
- The CrystalGenomics Drug Discovery Lab is supported by the BrainKorea21 (BK21) project. The BK21 project is a government driven funding system that aims to selectively support world-class research teams, and there are only 244 BK21 centers nation-wide. In particular, the CrystalGenomics Lab has published a cover paper in the Nature Vol. 425 (2003) for the discovery of the medicinal mechanism of Viagra. [^]
