(Matt Jacobson, Director; Andrej Sali and Brian Shoichet co-PIs)
A major goal of the EFI Computation Core is to develop strategies and tools for the prediction of enzyme function. With the UniProtKB/TrEMBL database recently surpassing 30,000,000 entries, it is more apparent than ever that computation is crucial to establishing this type of large scale strategy. Experimental characterization of every protein is not possible, nor even desirable, but many experimental hurdles, such as limited compound libraries and high cost but low throughput, can be circumvented through use of computational strategies. High quality functional predictions allow the community to begin to sift through the massive amount of genomic data now available and prioritize proteins with the most potential to yield interesting and relevant science.
Because enzymes and the systems in which they work are incredibly varied, computational approaches must be correspondingly varied to account for the unique chemistries and dynamics encountered. The Computation Core combines the skills of three accomplished labs headed by Matthew Jacobson, UCSF Department of Pharmaceutical Chemistry; Andrej Sali, UCSF Department of Bioengineering and Therapeutic Sciences; and Brian Shoichet, University of Toronto Faculty of Pharmacy (formally UCSF Department of Pharmaceutical Chemistry). Each lab brings unique expertise and perspectives to the challenge of computationally predicting enzymatic functions.
The Jacobson lab has been instrumental in developing methods to computationally evaluate, via modeling and docking, entire metabolic pathways. First benchmarked for E. coli glycolysis (Kalyanaraman and Jacobson 2010), the approach was recently used with great success to guide assignment of function for an Enolase (EN) Superfamily member initially structurally characterized by the Protein Structure Initiative (PDB 2PMQ). The Jacobson lab also implemented and tested a new integrated docking/rescoring algorithm that allows building ligands from a known position in an active site (e.g. from the diphosphate group in Isoprenoid Superfamily [IS] members). Use of this strategy enabled large scale analysis and prediction of function for the polyprenyl transferase subgroup of the IS Superfamily.
The Sali lab has deep experience in generation of automated homology models and widespread dissemination via their online tool, ModBase. New sequences are automatically processed weekly, and models are added to ModBase if a suitable template structure is available. Along with cross-linking ModBase to the Structure-Function Linkage Database (SFLD), a bioinformatic resource used in the EFI, all EFI targets are processed through MODPIPE. The models are updated every 6 months to reflect the newest available template structures. The Sali lab has adapted and benchmarked existing covalent docking methods for computational substrate and docking pose prediction for the Glutathione Transferase (GST) superfamily. The lab also worked with the Haloalkanoic Acid Dehalogenase (HAD) Bridging Project and Superfamily/Genome Core to develop structure-based multiple alignments for unique core domain structures.
The Shoichet lab has a long-standing history of approaching functional assignment in innovative fashions; for example, docking high energy intermediates (HEI) as a strategy for functional prediction in the amidohydrolase (AH) superfamily (Hermann et al. 2006). For the EFI, they have further applied this and other strategies to predict substrates for AH targets and are currently developing novel docking strategies for the HAD superfamily, which notably is characterized by generation of a covalent intermediate during the reaction mechanism.
In addition to their individual contributions, the computational labs routinely coordinate with each other, along with the other EFI groups, to address the specific needs of EFI superfamilies. For example, the Sali and Shoichet labs have joined forces repeatedly to attack challenges in the AH superfamily. Likewise, docking of covalently bound ligands has emerged as a critical need for the EFI; both Jacobson and Shoichet have developed these methods, which are being applied to the IS and HAD superfamilies, and the GST superfamily in collaboration.