Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (5 results)

  • Article
  • Peer Reviewed
UC San Francisco Previously Published Works (2012)
Benchmarks for molecular docking have historically focused on re-docking the cognate ligand of a well-determined protein-ligand complex to measure geometric pose prediction accuracy, and measurement of virtual screening performance has been focused on increasingly large and diverse sets of target protein structures, cognate ligands, and various types of decoy sets. Here, pose prediction is reported on the Astex Diverse set of 85 protein ligand complexes, and virtual screening performance is reported on the DUD set of 40 protein targets. In both cases, prepared structures of targets and ligands were provided by symposium organizers. The re-prepared data sets yielded results not significantly different than previous reports of Surflex-Dock on the two benchmarks. Minor changes to protein coordinates resulting from complex pre-optimization had large effects on observed performance, highlighting the limitations of cognate ligand re-docking for pose prediction assessment. Docking protocols developed for cross-docking, which address protein flexibility and produce discrete families of predicted poses, produced substantially better performance for pose prediction. Performance on virtual screening performance was shown to benefit by employing and combining multiple screening methods: docking, 2D molecular similarity, and 3D molecular similarity. In addition, use of multiple protein conformations significantly improved screening enrichment.
    • Thesis
    • Peer Reviewed
    UC San Francisco Electronic Theses and Dissertations (2013)

    Protein similarity has been used for the annotation and classification of proteins when the structure of the protein is available. Protein similarity comparisons may be made on a local or global basis and may consider sequence information and differing levels of structural information. This dissertation details the method Surflex PSIM, a local 3D method that compares the surfaces of protein binding sites.

    PSIM is a local 3D method that compares protein binding site surfaces in full atomic detail. The approach is based on the morphological similarity method (Surflex-Sim) which has been widely applied for global comparison of small molecules. This methodology has the ability to determine the differences between very similar proteins with different ligand binding specificity and the ability to correctly align extremely divergent proteins with only a small region of similarity. PSIM performed well on known standards for binding site comparisons.

    In a docking benchmark study, PSIM was used to assist in multi-structure docking protocols. In these protocols, proper selection of target structures can reduce time required for screening and increase accuracy. Selection of a minimal representative set of docking target conformations was performed automatically using PSIM. Several docking targets, for which unsatisfactory results had been obtained used a single-structure protocol, yielded substantial improvements using the PSIM-aided multi-structure docking protocol.

    Further development of an automated binding-site detection algorithm allowed for PSIM to be used as screening tool for annotating proteins with unknown function. A dataset was created of proteins whose function was determined after their crystallization. PSIM was able to automatically detect binding sites on a majority of these proteins and successfully match them to proteins that were present in the PDB at the time of crystallization that have the same function. PSIM was further used to explore possible functions for several proteins whose function is still unknown.

    The main contribution of this dissertation is a fast and accurate method for the comparison of protein binding sites agnostic of sequence information. This methodology has applications in the analysis of ligand specificity analysis and the annotation of proteins with unknown function.

      Cover page: Applications of a Surface-Based Protein Binding Site Comparison Methodology
      • Article
      • Peer Reviewed
      UC San Francisco Previously Published Works (2014)
      Hundreds of protein crystal structures exist for proteins whose function cannot be confidently determined from sequence similarity. Surflex-PSIM, a previously reported surface-based protein similarity algorithm, provides an alternative method for hypothesizing function for such proteins. The method now supports fully automatic binding site detection and is fast enough to screen comprehensive databases of protein binding sites. The binding site detection methodology was validated on apo/holo cognate protein pairs, correctly identifying 91% of ligand binding sites in holo structures and 88% in apo structures where corresponding sites existed. For correctly detected apo binding sites, the cognate holo site was the most similar binding site 87% of the time. PSIM was used to screen a set of proteins that had poorly characterized functions at the time of crystallization, but were later biochemically annotated. Using a fully automated protocol, this set of 8 proteins was screened against ∼60,000 ligand binding sites from the PDB. PSIM correctly identified functional matches that predated query protein biochemical annotation for five out of the eight query proteins. A panel of 12 currently unannotated proteins was also screened, resulting in a large number of statistically significant binding site matches, some of which suggest likely functions for the poorly characterized proteins.
        • Article
        • Peer Reviewed
        UC San Francisco Previously Published Works (2013)
        Binding affinity prediction is frequently addressed using computational models constructed solely with molecular structure and activity data. We present a hybrid structure-guided strategy that combines molecular similarity, docking, and multiple-instance learning such that information from protein structures can be used to inform models of structure-activity relationships. The Surflex-QMOD approach has been shown to produce accurate predictions of binding affinity by constructing an interpretable physical model of a binding site with no experimental binding site structural information. We introduce a method to integrate protein structure information into the model induction process in order to construct more robust physical models. The structure-guided models accurately predict binding affinities over a broad range of compounds while producing more accurate representations of the protein pockets and ligand binding modes. Structure-guidance for the QMOD method yielded significant performance improvements, both for affinity and pose prediction, especially in cases where predictions were made on ligands very different from those used for model induction.
          Cover page: A structure-guided approach for protein pocket modeling and affinity prediction.
          • Article
          • Peer Reviewed
          UC San Francisco Previously Published Works (2013)
          Binding affinity prediction is frequently addressed using computational models constructed solely with molecular structure and activity data. We present a hybrid structure-guided strategy that combines molecular similarity, docking, and multiple-instance learning such that information from protein structures can be used to inform models of structure-activity relationships. The Surflex-QMOD approach has been shown to produce accurate predictions of binding affinity by constructing an interpretable physical model of a binding site with no experimental binding site structural information. We introduce a method to integrate protein structure information into the model induction process in order to construct more robust physical models. The structure-guided models accurately predict binding affinities over a broad range of compounds while producing more accurate representations of the protein pockets and ligand binding modes. Structure-guidance for the QMOD method yielded significant performance improvements, both for affinity and pose prediction, especially in cases where predictions were made on ligands very different from those used for model induction.
            Top