|
|
|
|
|
|
| About Anthony E. Klon (Locus Pharmaceuticals) |
|
Dr. Anthony Klon received his B.S. in Biochemistry from the University of Washington in 1997. He pursued graduate studies in the Department of Biochemistry and Molecular Genetics under Professor Stephen C. Harvey at the University of Alabama at Birmingham where he received his PhD in 2002. His doctoral thesis focused on modeling the structure and dynamics of high-density lipoprotein particles, with an emphasis on the role of apolipoprotein A-I bound to discoidal HDL. During this time he also studied x-ray crystallography under Dr. David W. Borhani and solved structures of the ternary complexes of human dihydrofolate reductase with lipophilic antifolates. After graduation, he pursued his postdoctoral studies with Dr. John W. Davies at the Novartis Institutes for Biomedical Research in Cambridge, Massachusetts where he carried out an evaluation of high-throughput docking software against a variety of pharmaceutically relevant targets. While at Novartis, he developed novel approaches to improve the results of virtual high-throughput screening using machine learning algorithms and data fusion techniques. From 2005, he was a research scientist at Pharmacopeia in Princeton, New Jersey where he actively supported several drug discovery projects. He moved to Locus Pharmaceuticals in September 2008. His current research interests include the virtual screening of combinatorial libraries, dynamics simulations of protein-ligand complexes, and developing novel computational models for absorption, distribution, metabolism, excretion, and toxicity prediction.
|
|
Comparison of Machine Learning Algorithms to Predict ADME Properties Using Diverse Chemical Descriptors and Molecular Fingerprints
Anthony E. Klon and David J. Diller
We have compared the performance of ten different machine learning algorithms available in Weka to create binary classification models for blood-brain barrier (BBB) penetration and human intestinal absorption (HIA). For each data set, two models were constructed for each binary classifier; one using chemical descriptors and one using molecular fingerprints based on atom pairs and topological torsions, resulting in a total of 20 models for BBB penetration and HIA prediction. We describe the selection of descriptors used to train the chemical descriptor models. For both BBB and HIA datasets, the performance of all ten chemical descriptor models was tested by randomly scrambling the descriptors. For both datasets, the performance of all twenty models, descriptor and fingerprint-base, was further assessed and by randomly assigning compounds to the BBB penetrant / non-penetrant or HIA well-absorbed / poorly absorbed classes.
|
|
|
|
|
|
|
|
|
