Haiyuan Yu

Global landscape of protein complexes in the yeast Saccharomyces cerevisiae

Identification of protein–protein interactions often provides insight into protein function, and many cellular processes are performed by stable protein complexes. We used tandem affinity purification to process 4,562 different tagged proteins of the yeast Saccharomyces cerevisiae. Each preparation was analysed by both matrix-assisted laser desorption/ionization–time of flight mass spectrometry and liquid chromatography tandem mass spectrometry to increase coverage and accuracy. Machine learning was used to integrate the mass spectrometry scores and assign probabilities to the protein–protein interactions. Among 4,087 different proteins identified with high confidence by mass spectrometry from 2,357 successful purifications, our core data set (median precision of 0.69) comprises 7,123 protein–protein interactions involving 2,708 proteins. A Markov clustering algorithm organized these interactions into 547 protein complexes averaging 4.9 subunits per complex, about half of them absent from the MIPS database, as well as 429 additional interactions between pairs of complexes. The data (all of which are available online) will help future studies on individual proteins as well as functional genomics and systems biology.

High-quality binary protein interaction map of the yeast interactome network

Current yeast interactome network maps contain several hundred molecular complexes with limited and somewhat controversial representation of direct binary interactions. We carried out a comparative quality assessment of current yeast interactome data sets, demonstrating that high-throughput yeast two-hybrid (Y2H) screening provides high-quality binary interaction information. Because a large fraction of the yeast binary interactome remains to be mapped, we developed an empirically controlled mapping framework to produce a “second-generation” high-quality, high-throughput Y2H data set covering ∼20% of all yeast binary interactions. Both Y2H and affinity purification followed by mass spectrometry (AP/MS) data are of equally high quality but of a fundamentally different and complementary nature, resulting in networks with different topological and biological properties. Compared to co-complex interactome models, this binary map is enriched for transient signaling interactions and intercomplex connections with a highly significant clustering between essential proteins. Rather than correlating with essentiality, protein connectivity correlates with genetic pleiotropy.

Genomic analysis of regulatory network dynamics reveals large topological changes

Network analysis has been applied widely, providing a unifying language to describe disparate systems ranging from social interactions to power grids. It has recently been used in molecular biology, but so far the resulting networks have only been analysed statically. Here we present the dynamics of a biological network on a genomic scale, by integrating transcriptional regulatory information and gene-expression data for multiple conditions in Saccharomyces cerevisiae. We develop an approach for the statistical analysis of network dynamics, called SANDY, combining well-known global topological measures, local motifs and newly derived statistics. We uncover large changes in underlying network architecture that are unexpected given current viewpoints and random simulations. In response to diverse stimuli, transcription factors alter their interactions to varying degrees, thereby rewiring the network. A few transcription factors serve as permanent hubs, but most act transiently only during certain conditions. By studying sub-network structures, we show that environmental responses facilitate fast signal propagation (for example, with short regulatory cascades), whereas the cell cycle and sporulation direct temporal progression through multiple stages (for example, with highly inter-connected transcription factors). Indeed, to drive the latter processes forward, phase-specific transcription factors inter-regulate serially, and ubiquitously active transcription factors layer above them in a two-tiered hierarchy. We anticipate that many of the concepts presented here—particularly the large-scale topological changes and hub transience—will apply to other biological networks, including complex sub-systems in higher eukaryotes.

Detecting overlapping protein complexes in protein-protein interaction networks

We introduce clustering with overlapping neighborhood expansion (ClusterONE), a method for detecting potentially overlapping protein complexes from protein-protein interaction data. ClusterONE-derived complexes for several yeast data sets showed better correspondence with reference complexes in the Munich Information Center for Protein Sequence (MIPS) catalog and complexes derived from the Saccharomyces Genome Database (SGD) than the results of seven popular methods. The results also showed a high extent of functional homogeneity.

An experimentally derived confidence score for binary protein-protein interactions

Information on protein-protein interactions is of central importance for many areas of biomedical research. At present no method exists to systematically and experimentally assess the quality of individual interactions reported in interaction mapping experiments. To provide a standardized confidence-scoring method that can be applied to tens of thousands of protein interactions, we have developed an interaction tool kit consisting of four complementary, high-throughput protein interaction assays. We benchmarked these assays against positive and random reference sets consisting of well documented pairs of interacting human proteins and randomly chosen protein pairs, respectively. A logistic regression model was trained using the data from these reference sets to combine the assay outputs and calculate the probability that any newly identified interaction pair is a true biophysical interaction once it has been tested in the tool kit. This general approach will allow a systematic and empirical assignment of confidence scores to all individual protein-protein interactions in interactome networks.

Three-dimensional reconstruction of protein networks provides insight into human genetic disease

To better understand the molecular mechanisms and genetic basis of human disease, we systematically examine relationships between 3,949 genes, 62,663 mutations and 3,453 associated disorders by generating a three-dimensional, structurally resolved human interactome. This network consists of 4,222 high-quality binary protein-protein interactions with their atomic-resolution interfaces. We find that in-frame mutations (missense point mutations and in-frame insertions and deletions) are enriched on the interaction interfaces of proteins associated with the corresponding disorders, and that the disease specificity for different mutations of the same gene can be explained by their location within an interface. We also predict 292 candidate genes for 694 unknown disease-to-gene associations with proposed molecular mechanism hypotheses. This work indicates that knowledge of how in-frame disease mutations alter specific interactions is critical to understanding pathogenesis. Structurally resolved interaction networks should be valuable tools for interpreting the wealth of data being generated by large-scale structural genomics and disease association studies.

Genomic analysis of the hierarchical structure of regulatory networks

A fundamental question in biology is how the cell uses transcription factors (TFs) to coordinate the expression of thousands of genes in response to various stimuli. The relationships between TFs and their target genes can be modeled in terms of directed regulatory networks. These relationships, in turn, can be readily compared with commonplace “chain-of-command” structures in social networks, which have characteristic hierarchical layouts. Here, we develop algorithms for identifying generalized hierarchies (allowing for various loop structures) and use these approaches to illuminate extensive pyramid-shaped hierarchical structures existing in the regulatory networks of representative prokaryotes (Escherichia coli) and eukaryotes (Saccharomyces cerevisiae), with most TFs at the bottom levels and only a few master TFs on top. These masters are situated near the center of the protein–protein interaction network, a different type of network from the regulatory one, and they receive most of the input for the whole regulatory hierarchy through protein interactions. Moreover, they have maximal influence over other genes, in terms of affecting expression-level changes. Surprisingly, however, TFs at the bottom of the regulatory hierarchy are more essential to the viability of the cell. Finally, one might think master TFs achieve their wide influence through directly regulating many targets, but TFs with most direct targets are in the middle of the hierarchy. We find, in fact, that these midlevel TFs are “control bottlenecks” in the hierarchy, and this great degree of control for “middle managers” has parallels in efficient social structures in various corporate and governmental settings.

Edgetic perturbation models of human inherited disorders

Cellular functions are mediated through complex systems of macromolecules and metabolites linked through biochemical and physical interactions, represented in interactome models as ‘nodes’ and ‘edges’, respectively. Better understanding of genotype‐to‐phenotype relationships in human disease will require modeling of how disease‐causing mutations affect systems or interactome properties. Here we investigate how perturbations of interactome networks may differ between complete loss of gene products (‘node removal’) and interaction‐specific or edge‐specific (‘edgetic’) alterations. Global computational analyses of ∼50 000 known causative mutations in human Mendelian disorders revealed clear separations of mutations probably corresponding to those of node removal versus edgetic perturbations. Experimental characterization of mutant alleles in various disorders identified diverse edgetic interaction profiles of mutant proteins, which correlated with distinct structural properties of disease proteins and disease mechanisms. Edgetic perturbations seem to confer distinct functional consequences from node removal because a large fraction of cases in which a single gene is linked to multiple disorders can be modeled by distinguishing edgetic network perturbations. Edgetic network perturbation models might improve both the understanding of dissemination of disease alleles in human populations and the development of molecular therapeutic strategies.

Normal mode analysis of macromolecular motions in a database framework: Developing mode concentration as a useful classifying statistic

We investigated protein motions using normal modes within a database framework, determining on a large sample the degree to which normal modes anticipate the direction of the observed motion and were useful for motions classification. As a starting point for our analysis, we identified a large number of examples of protein flexibility from a comprehensive set of structural alignments of the proteins in the PDB. Each example consisted of a pair of proteins that were considerably different in structure given their sequence similarity. On each pair, we performed geometric comparisons and adiabatic‐mapping interpolations in a high‐throughput pipeline, arriving at a final list of 3,814 putative motions and standardized statistics for each. We then computed the normal modes of each motion in this list, determining the linear combination of modes that best approximated the direction of the observed motion. We integrated our new motions and normal mode calculations in the Macromolecular Motions Database, through a new ranking interface at http://molmovdb.org. Based on the normal mode calculations and the interpolations, we identified a new statistic, mode concentration, related to the mathematical concept of information content, which describes the degree to which the direction of the observed motion can be summarized by a few modes. Using this statistic, we were able to determine the fraction of the 3,814 motions where one could anticipate the direction of the actual motion from only a few modes. We also investigated mode concentration in comparison to related statistics on combinations of normal modes and correlated it with quantities characterizing protein flexibility (e.g., maximum backbone displacement or number of mobile atoms). Finally, we evaluated the ability of mode concentration to automatically classify motions into a variety of simple categories (e.g., whether or not they are “fragment‐like”), in comparison to motion statistics. This involved the application of decision trees and feature selection (particular machine‐learning techniques) to training and testing sets derived from merging the “list” of motions with manually classified ones. Proteins 2002;48:682–695.

Integrative annotation of variants from 1092 humans: application to cancer genomics.

Introduction Plummeting sequencing costs have led to a great increase in the number of personal genomes. Interpreting the large number of variants in them, particularly in noncoding regions, is a current challenge. This is especially the case for somatic variants in cancer genomes, a large proportion of which are noncoding. Prioritization of candidate noncoding cancer drivers based on patterns of selection. (Step 1) Filter somatic variants to exclude 1000 Genomes polymorphisms; (2) retain variants in noncoding annotations; (3) retain those in “sensitive” regions; (4) prioritize those disrupting a transcription-factor binding motif and (5) residing near the center of a biological network; (6) prioritize ones in annotation blocks mutated in multiple cancer samples. Methods We investigated patterns of selection in DNA elements from the ENCODE project using the full spectrum of variants from 1092 individuals in the 1000 Genomes Project (Phase 1), including single-nucleotide variants (SNVs), short insertions and deletions (indels), and structural variants (SVs). Although we analyzed broad functional annotations, such as all transcription-factor binding sites, we focused more on highly specific categories such as distal binding sites of factor ZNF274. The greater statistical power of the Phase 1 data set compared with earlier ones allowed us to differentiate the selective constraints on these categories. We also used connectivity information between elements from protein-protein-interaction and regulatory networks. We integrated all the information on selection to develop a workflow (FunSeq) to prioritize personal-genome variants on the basis of their deleterious impact. As a proof of principle, we experimentally validated and characterized a few candidate variants. Results We identified a specific subgroup of noncoding categories with almost as much selective constraint as coding genes: “ultrasensitive” regions. We also uncovered a number of clear patterns of selection. Elements more consistently active across tissues and both maternal and paternal alleles (in terms of allele-specific activity) are under stronger selection. Variants disruptive because of mechanistic effects on transcription-factor binding (i.e. “motif-breakers”) are selected against. Higher network connectivity (i.e. for hubs) is associated with higher constraint. Additionally, many hub promoters and regulatory elements show evidence of recent positive selection. Overall, indels and SVs follow the same pattern as SNVs; however, there are notable exceptions. For instance, enhancers are enriched for SVs formed by nonallelic homologous recombination. We integrated these patterns of selection into the FunSeq prioritization workflow and applied it to cancer variants, because they present a strong contrast to inherited polymorphisms. In particular, application to ~90 cancer genomes (breast, prostate and medulloblastoma) reveals nearly a hundred candidate noncoding drivers. Discussion Our approach can be readily used to prioritize variants in cancer and is immediately applicable in a precision-medicine context. It can be further improved by incorporation of larger-scale population sequencing, better annotations, and expression data from large cohorts. Identifying Important Identifiers Each of us has millions of sequence variations in our genomes. Signatures of purifying or negative selection should help identify which of those variations is functionally important. Khurana et al. (1235587) used sequence polymorphisms from 1092 humans across 14 populations to identify patterns of selection, especially in noncoding regulatory regions. Noncoding regions under very strong negative selection included binding sites of some chromatin and general transcription factors (TFs) and core motifs of some important TF families. Positive selection in TF binding sites tended to occur in network hub promoters. Many recurrent somatic cancer variants occurred in noncoding regulatory regions and thus might indicate mutations that drive cancer. Regions under strong selection in the human genome identify noncoding regulatory elements with possible roles in disease. Interpreting variants, especially noncoding ones, in the increasing number of personal genomes is challenging. We used patterns of polymorphisms in functionally annotated regions in 1092 humans to identify deleterious variants; then we experimentally validated candidates. We analyzed both coding and noncoding regions, with the former corroborating the latter. We found regions particularly sensitive to mutations (“ultrasensitive”) and variants that are disruptive because of mechanistic effects on transcription-factor binding (that is, “motif-breakers”). We also found variants in regions with higher network centrality tend to be deleterious. Insertions and deletions followed a similar pattern to single-nucleotide variants, with some notable exceptions (e.g., certain deletions and enhancers). On the basis of these patterns, we developed a computational tool (FunSeq), whose application to ~90 cancer genomes reveals nearly a hundred candidate noncoding drivers.