Information Bottleneck and Aggregated Learning.

We consider the problem of learning a neural network classifier. Under the information bottleneck (IB) principle, we associate with this classification problem a representation learning problem, which we call "IB learning". We show that IB learning is, in fact, equivalent to a special class of the quantization problem. The classical results in rate-distortion theory then suggest that IB learning can benefit from a "vector quantization" approach, namely, simultaneously learning the representations of multiple input objects. Such an approach assisted with some variational techniques, result in a novel learning framework, "Aggregated Learning", for classification with neural network models. In this framework, several objects are jointly classified by a single neural network. The effectiveness of this framework is verified through extensive experiments on standard image recognition and text classification tasks.

Evolutionary versatility of eukaryotic protein domains revealed by their bigram networks.

BACKGROUND: Protein domains are globular structures of independently folded polypeptides that exert catalytic or binding activities. Their sequences are recognized as evolutionary units that, through genome recombination, constitute protein repertoires of linkage patterns. Via mutations, domains acquire modified functions that contribute to the fitness of cells and organisms. Recent studies have addressed the evolutionary selection that may have shaped the functions of individual domains and the emergence of particular domain combinations, which led to new cellular functions in multi-cellular animals. This study focuses on modeling domain linkage globally and investigates evolutionary implications that may be revealed by novel computational analysis. RESULTS: A survey of 77 completely sequenced eukaryotic genomes implies a potential hierarchical and modular organization of biological functions in most living organisms. Domains in a genome or multiple genomes are modeled as a network of hetero-duplex covalent linkages, termed bigrams. A novel computational technique is introduced to decompose such networks, whereby the notion of domain "networking versatility" is derived and measured. The most and least "versatile" domains (termed "core domains" and "peripheral domains" respectively) are examined both computationally via sequence conservation measures and experimentally using selected domains. Our study suggests that such a versatility measure extracted from the bigram networks correlates with the adaptivity of domains during evolution, where the network core domains are highly adaptive, significantly contrasting the network peripheral domains. CONCLUSIONS: Domain recombination has played a major part in the evolution of eukaryotes attributing to genome complexity. From a system point of view, as the results of selection and constant refinement, networks of domain linkage are structured in a hierarchical modular fashion. Domains with high degree of networking versatility appear to be evolutionary adaptive, potentially through functional innovations. Domain bigram networks are informative as a model of biological functions. The networking versatility indices extracted from such networks for individual domains reflect the strength of evolutionary selection that the domains have experienced.

Eukaryotic protein domains as functional units of cellular evolution.

Modular protein domains are functional units that can be modified through the acquisition of new intrinsic activities or by the formation of novel domain combinations, thereby contributing to the evolution of proteins with new biological properties. Here, we assign proteins to groups with related domain compositions and functional properties, termed "domain clubs," which we use to compare multiple eukaryotic proteomes. This analysis shows that different domain types can take distinct evolutionary trajectories, which correlate with the conservation, gain, expansion, or decay of particular biological processes. Evolutionary jumps are associated with a domain that coordinately acquires a new intrinsic function and enters new domain clubs, thereby providing the modified domain with access to a new cellular microenvironment. We also coordinately analyzed the covalent and noncovalent interactions of different domain types to assess the molecular compartment occupied by each domain. This reveals that specific subsets of domains demarcate particular cellular processes, such as growth factor signaling, chromatin remodeling, apoptotic and inflammatory responses, or vesicular trafficking. We suggest that domains, and the proteins in which they reside, are selected during evolution through reciprocal interactions with protein domains in their local microenvironment. Based on this scheme, we propose a mechanism by which Tudor domains may have evolved to support different modes of epigenetic regulation and suggest a role for the germline group of mammalian Tudor domains in Piwi-regulated RNA biology.

Informatics platform for global proteomic profiling and biomarker discovery using liquid chromatography-tandem mass spectrometry.

We have developed an integrated suite of algorithms, statistical methods, and computer applications to support large-scale LC-MS-based gel-free shotgun profiling of complex protein mixtures using basic experimental procedures. The programs automatically detect and quantify large numbers of peptide peaks in feature-rich ion mass chromatograms, compensate for spurious fluctuations in peptide signal intensities and retention times, and reliably match related peaks across many different datasets. Application of this toolkit markedly facilitates pattern recognition and biomarker discovery in global comparative proteomic studies, simplifying mechanistic investigation of physiological responses and the detection of proteomic signatures of disease.