The broad goal is to develop computational techniques to study protein-ligand interactions. The developed methods are applied to GPCRs, transporters, and enzymes (e.g. kinases), to contribute to a better understanding and regulation of biological processes, as well as the discovery of new therapeutics. The computational predictions are tested experimentally through collaborations.
We are developing computational techniques to effectively and accurately model protein-ligand interactions. For instance, we constructed an in-silico platform for chemical toxicity prediction. Using toxicity data of 10 nuclear receptors from ToxCAST, we systematically analyzed how multiple crystal protein structures can be used in screening protocols to distinguish actives from inactives. The methods used for data fusion are consensus docking scores from multiple conformations at a single functional state (agonist-bound or antagonist-bound), multiple functional states, and multiple pockets (orthosteric site and allosteric sites). The results suggested that such a consensus protocol could have a greater discriminating power between active and inactive molecules with respect to the standard single-structure based approach. The complete workflow is summarized in Figure 1.
GPCRs play important roles in cell signaling pathways. Their dysfunction causes many human developmental and metabolic disorders, as well as certain cancers. However, being a membrane protein, it is difficult to obtain the 3D crystal structures of GPCRs for virtual screening of ligands by molecular docking. We evaluated the virtual screening performance of homology models of 19 human GPCRs with respect to the corresponding crystal structures, suggesting that ligand candidates predicted with consensus scores from multiple models can be the optimal option in practical applications where the performance of each model cannot be estimated. Meanwhile, in collaboration with Dr. Cheng Zhang from the University of Pittsburgh, we contributed to the determination of crystal structures of GPCRs including a lipid GPCR CRTH2 as the receptor for prostaglandin D2, the C5a receptor (C5aR) that can induce strong inflammatory events in response to the anaphylatoxin C5a peptide, and N-formyl peptide receptor 2 (FPR2) that is a high affinity receptor for the arachidonic acid metabolites. Currently we are working towards the identification of novel orthosteric and allosteric modulators and better understanding of structure-function mechanisms, of these GPCRs using in-silico approaches.
We are collaborating with BII, ICES, and IHPC colleagues, developing “an Integrative Computational/Experimental Workflow for Enzyme Development”, to improve the speed of enzyme engineering and to build a panel of enzymes with activity against industrially relevant panel of substrates. Galactose oxidase (GAOx) from copper-radical oxidases (CROs) was selected as the 1st enzyme class to focus on, as it catalyzes a range of primary alcohols such as alcohols, glycerols, and sugars to corresponding aldehydes and ketones. Going forward, we will employ protein structure modeling, molecular docking, and machine learning approaches to predict the catalytic activity of the redesigned variants (Figure 4). ).
In the meantime, we also collaborate with Dr. Wen Shan Yew from Department of Biochemistry, NUS, to develop bacteria lactonase and prenyltransferases. The lactonase catalyzes the quorum sensing signal N-acyl homoserine lactones (AHLs), as antivirulence therapeutic agents. We start from a thermostable GKL (lactonase from Geobacillus kaustophilus) enzyme, suggest mutations in the enzyme active site by computational approaches, and test these mutations in enzyme functional assays. Some mutations show increased catalytic efficiencies and a broadened substrate range. Prenyltransferases are enzymes that attach prenyl groups to other molecules. This building capability is important in many natural product synthetic pathways, such as cannabinoid biosynthesis. We are developing prenyltransferases that are catalytically more efficient and with higher substrate diversity by rational mutagenesis.
High-throughput (HTP) models are required in order to assess the risks posed by the large number of existing or novel xenobiotics used in pharmaceutical, food, personal care, agricultural, chemical and other industries. We are developing in-silico methods to help the identification of the mode-of-action (MoA) of xenobiotics. The in-silico methods are currently evaluated using two family of proteins: nuclear receptors and cytochrome P450 (CYP) enzymes. Many xenobiotics exert their harmful effects by resembling endogenous hormones and falsely activating nuclear receptors (NRs). We are evaluating the utility of our in-silico methods to distinguish active and inactive chemicals for NRs. Some predictions have been tested in vitro. The family of CYP enzymes plays an important role in the metabolism of a large number of endogenous and exogenous compounds, including most of the drugs currently on the market.
We collaborate with Dr. Lit-Hsin Loo in BII, A*STAR and Dr. Eric Chan in Department of Pharmacy, NUS, benchmarking and further improving existing in-silico methods, to facilitate CYP-mediated ToxCAST actives and inactives (Figure 5). These projects will provide important basis for the in-silico and in-vitro tissue or multi-organ toxicity models that we aim to develop in the near future.
Ebola (EBOV) and Marburg (MARV) are members of the Filoviridae family, which continue to emerge and cause sporadic outbreaks of hemorrhagic fever with high mortality rates. Filoviruses utilize their VP40 viral matrix protein to drive virion assembly and budding, in part, by recruitment of specific WW domain-bearing host proteins via its conserved PPxY domain motif. We have docked five PPxY-containing peptides from Marburg (MP), Ebola (EP) and angiomotin (AMOT) peptides (AP1-3) to Yes Associated Protein (YAP), as well as a control peptide, to identify and understand the virus-host interactions (Figure 6). Our protein-peptide docking results show PPxY motif of all the five peptides bind to the PPxY binding region of YAP WW1 domain. MP shows a better docking score compared to other peptides, followed by AP2, AP1, EP and AP3 peptide. These observations indicate that MP may bind stronger compared to other peptides.
Dr. Hao Fan received his undergraduate degree in Biological Sciences in University of Science and Technology of China (USTC). He obtained his Ph.D in Biophysical Chemistry in Dr. Alan Mark's lab at University of Groningen (RUG). He worked as postdoctoral fellow followed by research scientist in both Dr. Andrej Sali's lab and Dr. Brian Shoichet's lab at University of California, San Francisco (UCSF). He was appointed Principal Investigator at the Bioinformatics Institute (BII), A*STAR in February 2014. The broad goal of Fan Lab is to develop computational techniques for modeling protein-ligand interactions, and to apply the developed methods in applications including:
The broad goal is to develop computational techniques to effectively and accurately model protein-ligand interactions, the developed methods will be applied to therapeutic targets such as GPCRs, transporters and downstream kinases, to contribute to a better understanding and regulation of biological processes, to the discovery of new ingredients for food and nutrition, chemical probes, and drug leads, and to the development of an in-silico platform for chemical toxicity prediction.