Duncan Penfold-Brown

Parametric Bayesian priors and better choice of negative examples improve protein function prediction

MOTIVATION Computational biologists have demonstrated the utility of using machine learning methods to predict protein function from an integration of multiple genome-wide data types. Yet, even the best performing function prediction algorithms rely on heuristics for important components of the algorithm, such as choosing negative examples (proteins without a given function) or determining key parameters. The improper choice of negative examples, in particular, can hamper the accuracy of protein function prediction. RESULTS We present a novel approach for choosing negative examples, using a parameterizable Bayesian prior computed from all observed annotation data, which also generates priors used during function prediction. We incorporate this new method into the GeneMANIA function prediction algorithm and demonstrate improved accuracy of our algorithm over current top-performing function prediction methods on the yeast and mouse proteomes across all metrics tested. AVAILABILITY Code and Data are available at: http://bonneaulab.bio.nyu.edu/funcprop.html

Negative example selection for protein function prediction: the NoGO database.

Negative examples – genes that are known not to carry out a given protein function – are rarely recorded in genome and proteome annotation databases, such as the Gene Ontology database. Negative examples are required, however, for several of the most powerful machine learning methods for integrative protein function prediction. Most protein function prediction efforts have relied on a variety of heuristics for the choice of negative examples. Determining the accuracy of methods for negative example prediction is itself a non-trivial task, given that the Open World Assumption as applied to gene annotations rules out many traditional validation metrics. We present a rigorous comparison of these heuristics, utilizing a temporal holdout, and a novel evaluation strategy for negative examples. We add to this comparison several algorithms adapted from Positive-Unlabeled learning scenarios in text-classification, which are the current state of the art methods for generating negative examples in low-density annotation contexts. Lastly, we present two novel algorithms of our own construction, one based on empirical conditional probability, and the other using topic modeling applied to genes and annotations. We demonstrate that our algorithms achieve significantly fewer incorrect negative example predictions than the current state of the art, using multiple benchmarks covering multiple organisms. Our methods may be applied to generate negative examples for any type of method that deals with protein function, and to this end we provide a database of negative examples in several well-studied organisms, for general use (The NoGO database, available at: bonneaulab.bio.nyu.edu/nogo.html).

The Plant Proteome Folding Project: Structure and Positive Selection in Plant Protein Families

Despite its importance, relatively little is known about the relationship between the structure, function, and evolution of proteins, particularly in land plant species. We have developed a database with predicted protein domains for five plant proteomes (http://pfp.bio.nyu.edu) and used both protein structural fold recognition and de novo Rosetta-based protein structure prediction to predict protein structure for Arabidopsis and rice proteins. Based on sequence similarity, we have identified ∼15,000 orthologous/paralogous protein family clusters among these species and used codon-based models to predict positive selection in protein evolution within 175 of these sequence clusters. Our results show that codons that display positive selection appear to be less frequent in helical and strand regions and are overrepresented in amino acid residues that are associated with a change in protein secondary structure. Like in other organisms, disordered protein regions also appear to have more selected sites. Structural information provides new functional insights into specific plant proteins and allows us to map positively selected amino acid sites onto protein structures and view these sites in a structural and functional context.

Turning the Virtual Tables: Government Strategies for Addressing Online Opposition with an Application to Russia

We introduce a novel classification of strategies employed by autocrats to combat hostile activity on the web and in social media in particular. Our classification looks at these options from the point of view of the end internet user and distinguishes both online from offline response and exerting control from engaging in opinion formation. For each of the three options – offline action, infrastructure regulation and online engagement – we provide a detailed account for the evolution of Russian government strategy since 2000. In addition, for online engagement option we construct the tools for detecting such activity on Twitter and test them on a large dataset of politically relevant Twitter data from ∗Corresponding author: sanovich@nyu.edu.

Big data, social media, and protest: foundations for a research agenda

INTRODUCTION The past decade has witnessed a rapid rise in the use of social media around the globe. For political scientists, this is a phenomenon begging to be understood. It has been claimed repeatedly – usually in the absence of solid data – that these social media resources are profoundly shaping participation in social movements, including protest movements (see Bond, Fariss, Jones, Kramer, Marlow, Settle, & Fowler 2012; Cha et al. 2010; Jungherr, Jurgens, & Schoen 2012; Lynch 2011; Shirky 2011). Social media are often assumed to affect an extremely wide range of individual-level behaviors, including communicating about politics to friends and family members, donating or soliciting money for political campaigns and causes, voting, and engaging in collective forms of protest. In truth, however, the research community knows remarkably little about whether ( and especially how ) the use of social media systematically affects political participation. Perhaps nowhere is this lack of knowledge more clear than in the matter of political protest. In recent years, the use of social media has been linked to the spread of political protests in cities around the world, including Moscow, Kiev, Istanbul, Ankara, Cairo, Tripoli, Athens, Madrid, New York, and Los Angeles. Obviously, social protest itself is far from new, but the fact that it is possible for potential protest participants, as well as geographically removed observers, to access real-time accounts of protest behavior documented and archived through micro-blogging (e.g., Twitter) and social media (e.g., Facebook) websites is a novel phenomenon. Protest activities are flagged by participants themselves with distinctive hashtags on Twitter. As political scientists, then, the question of how these activities on social media actually affect the decision of individuals to participate in protests would seem to be a subject ripe for research, as too is the macro question of how social media changes the nature of protest itself. As is often the case in both popular and scholarly commentary, new phenomena inevitably engender a counternarrative, claiming that the phenomenon is either not new or not important. The rise of social media – and the concurrent level of fascination accorded to it across the media spectra – and its relationship to mass protests movements have been no exception in this regard.

Digital Dissent: An Analysis of the Motivational Contents of Tweets From an Occupy Wall Street Demonstration.

Social scientific models of protest activity emphasize instrumental motives associated with rational self-interest and beliefs about group efficacy and symbolic motives associated with social identification and anger at perceived injustice. Ideological processes are typically neglected, despite the fact that protest movements occur in a sociopolitical context in which some people are motivated to maintain the status quo, whereas others are motivated to challenge it. To investigate the role of ideology and other social psychological processes in protest participation, we used manual and machine-learning methods to analyze the contents of 23,810 tweets sent on the day of the May Day 2012 Occupy Wall Street demonstration along with an additional 664,937 tweets (sent by 8,244 unique users) during the 2-week lead-up to the demonstration. Results revealed that social identification and liberal ideology were significant independent predictors of protest participation. The effect of social identification was mediated by the expression of collective efficacy, justice concerns, ideological themes, and positive emotion. The effect of liberalism was mediated by the expression of ideological themes, but conservatives were more likely to express ideological backlash against Occupy Wall Street than liberals were to express ideological support for the movement or demonstration. The expression of self-interest and anger was either negatively related or unrelated to protest participation. This work illustrates the promise (and challenge) of using automated methods to analyze new, ecologically valid data sources for studying protest activity and its motivational underpinnings—thereby informing strategic campaigns that employ collective action tactics.