Epik version 7 is a software program that uses machine learning for predicting the pK a values and protonation state distribution of complex, druglike molecules. Using an ensemble of atomic graph convolutional neural networks (GCNNs) trained on over 42,000 pK a values across broad chemical space from both experimental and computed origins, the model predicts pK a values with 0.42 and 0.72 pK a unit median absolute and root mean square errors, respectively, across seven test sets. Epik version 7 also generates protonation states and recovers 95% of the most populated protonation states compared to previous versions. Requiring on average only 47 ms per ligand, Epik version 7 is rapid and accurate enough to evaluate protonation states for crucial molecules and prepare ultra-large libraries of compounds to explore vast regions of chemical space. The simplicity and time required for the training allow for the generation of highly accurate models customized to a program's specific chemistry.

Dihydroorotate dehydrogenase (DHODH) has been clinically validated as a target for the development of new antimalarials. Experience with clinical candidate triazolopyrimidine DSM265 (1) suggested that DHODH inhibitors have great potential for use in prophylaxis, which represents an unmet need in the malaria drug discovery portfolio for endemic countries, particularly in areas of high transmission in Africa. We describe a structure-based computationally driven lead optimization program of a pyrrole-based series of DHODH inhibitors, leading to the discovery of two candidates for potential advancement to preclinical development. These compounds have improved physicochemical properties over prior series frontrunners and they show no timedependent CYP inhibition, characteristic of earlier compounds. Frontrunners have potent antimalarial activity in vitro against blood and liver schizont stages and show good efficacy in Plasmodium falciparum SCID mouse models. They are equally active against P. falciparum and Plasmodium vivax field isolates and are selective for Plasmodium DHODHs versus mammalian enzymes.

Epik version 7 is a software program that uses machine learning for predicting the pKa values and protonation state distribution of complex, drug-like molecules. Using an ensemble of atomic graph convolutional neural networks (GCNNs) trained on over 42,000 pKa values across broad chemical space from both experimental and computed origins, the model predicts pKa values with 0.42 and 0.72 log unit median absolute and RMS errors, respectively, across seven test sets. Epik version 7 also generates protonation states and recovers 95% of the most populated protonation states compared to previous versions. Requiring on average only 47 ms per ligand, Epik version 7 is rapid and accurate enough to evaluate protonation states for crucial molecules and prepare ultra-large libraries of compounds to explore vast regions of chemical space. The simplicity of and time required for the training allows for the generation of highly accurate models customized to a program’s specific chemistry.

MDM2 and MDMX function as key regulators of p53 by binding to its N terminus, inhibiting its transcriptional activity, and promoting its degradation. In particular, MDM2 is overexpressed in some of human tumors, and with MDMX contributes directly to loss of p53 function during the development of nearly 50% of human cancers. Due to p53 inactivation, MDM2 in many tumors confers tumor survival; therefore it is an important molecular target for anticancer therapy. Several studies showed that reactivation of wild type p53 in tumor cells can be obtained by disrupting the MDM2/p53 interaction with peptidic, peptidomimetic, and small molecule p53-mimetics. Specific successful examples include the Nutlins and spirooxindole analogs (MI-219 and MI-63). Amongst the peptidic and peptidomimetic inhibitors examined to date, none is nearly as effective as Nutlins and MI-219 in tumor killing in vitro. Hence, new inhibitors against MDM2 and/or MDMX are needed: as cell permeable chemical probes of the p53 pathway in cancer biology, and as templates for structure-based rational design of p53 activators for future therapeutic use. As part of our drug discovery program to identify antagonists of the p53/MDM2 and p53/MDMx protein-protein interactions, a high-throughput in-silico screen of a 3.2 millions virtual library of compounds (from Schrödinger, Inc.). A physical restraint was applied during the screen, in order to mimic binding to the hydrophobic cleft of MDM2 normally occupied by three p53 side chains (F19, W23, and L26) that are critical for MDM2/p53 binding. The top highest ranked 160 compounds were then assessed for their ability to block p53 interaction with MDM2 and MDMx in an ELISA assay. This resulted in the identification of E12/DP3-117, a small molecule disruptor of the p53/MDM2 protein-protein interaction with an IC50 value of 47 ± 14 μM. We will report the synthesis and biological evaluation of focused libraries based on the initial hit and on compounds showing improved activity. Structure activity relationship studies around the hits will be disclosed as well as the outcomes of further rounds of chemical design and biological assessment. Binding of E12/DP3-117 to MDM2 is currently being assessed via co-crystallization and other biophysical techniques. We will describe the use of the crystal structure of p53-like mutant peptides in complex with the N-terminal domains of Mdm2, as the basis for rational design of more potent MDM2 small-molecule/peptide hybrid inhibitors. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 3242. doi:10.1158/1538-7445.AM2011-3242

Epik version 7 is a software program that uses machine learning for predicting the pKa values and protonation state distribution of complex, drug-like molecules. Using an ensemble of atomic graph convolutional neural networks (GCNNs) trained on over 42,000 pKa values across broad chemical space from both experimental and computed origins, the model predicts pKa values with 0.42 and 0.72 log unit median absolute and RMS errors, respectively, across seven test sets. Epik version 7 also generates protonation states and recovers 95% of the most populated protonation states compared to previous versions. Requiring on average only 47 ms per ligand, Epik version 7 is rapid and accurate enough to evaluate protonation states for crucial molecules and prepare ultra-large libraries of compounds to explore vast regions of chemical space. The simplicity of and time required for the training allows for the generation of highly accurate models customized to a program’s specific chemistry.