Context-guided entropy minimization for semi-supervised domain adaptation.

Semi-Supervised Domain Adaptation has been widely studied with various approaches to address domain shift with labeled source-domain data combined with scarcely labeled target-domain data. Model adaptation is becoming promising with a paradigm of source pre-training and target fine-tuning, which eliminates the simultaneous availability of data from both domains and makes for data privacy. Among the model adaptation methods, Entropy Minimization (EM) is popularly incorporated to encourage a low-density separation on target samples. However, EM tends to brutally force models to make over-confident predictions, which could make the models collapse with deteriorated performance. In this paper, we first study the over-confidence of EM with a quantitative analysis, which shows the importance of capturing the dependency among labels. To address this issue, we propose to guide EM via longitudinal self-distillation. Specifically, we produce a dynamic "teacher" label distribution during training by constructing a graph on target data and perform pseudo-label propagation to encourage the "teacher" distribution to capture context category dependency based on a global data structure. Then EM is guided longitudinally by distilling the learned label distribution to combat the brute-force over-confidence. Extensive experiments demonstrate the effectiveness of our methods.

In vitro validation of in silico identified inhibitory interactions.

BACKGROUND: Understanding how neuronal signals propagate in local network is an important step in understanding information processing. As a result, spike trains recorded with multi-electrode arrays (MEAs) have been widely used to study the function of neural networks. Studying the dynamics of neuronal networks requires the identification of both excitatory and inhibitory connections. The detection of excitatory relationships can robustly be inferred by characterizing the statistical relationships of neural spike trains. However, the identification of inhibitory relationships is more difficult: distinguishing endogenous low firing rates from active inhibition is not obvious. NEW METHOD: In this paper, we propose an in silico interventional procedure that makes predictions about the effect of stimulating or inhibiting single neurons on other neurons, and thereby gives the ability to accurately identify inhibitory effects. COMPARISON: To experimentally test these predictions, we have developed a Neural Circuit Probe (NCP) that delivers drugs transiently and reversibly on individually identified neurons to assess their contributions to the neural circuit behavior. RESULTS: Using the NCP, putative inhibitory connections identified by the in silico procedure were validated through in vitro interventional experiments. CONCLUSIONS: Together, these results demonstrate how detailed microcircuitry can be inferred from statistical models derived from neurophysiology data.

The genome of the jellyfish Aurelia and the evolution of animal complexity.

We present the genome of the moon jellyfish Aurelia, a genome from a cnidarian with a medusa life stage. Our analyses suggest that gene gain and loss in Aurelia is comparable to what has been found in its morphologically simpler relatives-the anthozoan corals and sea anemones. RNA sequencing analysis does not support the hypothesis that taxonomically restricted (orphan) genes play an oversized role in the development of the medusa stage. Instead, genes broadly conserved across animals and eukaryotes play comparable roles throughout the life cycle. All life stages of Aurelia are significantly enriched in the expression of genes that are hypothesized to interact in protein networks found in bilaterian animals. Collectively, our results suggest that increased life cycle complexity in Aurelia does not correlate with an increased number of genes. This leads to two possible evolutionary scenarios: either medusozoans evolved their complex medusa life stage (with concomitant shifts into new ecological niches) primarily by re-working genetic pathways already present in the last common ancestor of cnidarians, or the earliest cnidarians had a medusa life stage, which was subsequently lost in the anthozoans. While we favour the earlier hypothesis, the latter is consistent with growing evidence that many of the earliest animals were more physically complex than previously hypothesized.

Query-Based Outlier Detection in Heterogeneous Information Networks.

Outlier or anomaly detection in large data sets is a fundamental task in data science, with broad applications. However, in real data sets with high-dimensional space, most outliers are hidden in certain dimensional combinations and are relative to a user's search space and interest. It is often more effective to give power to users and allow them to specify outlier queries flexibly, and the system will then process such mining queries efficiently. In this study, we introduce the concept of query-based outlier in heterogeneous information networks, design a query language to facilitate users to specify such queries flexibly, define a good outlier measure in heterogeneous networks, and study how to process outlier queries efficiently in large data sets. Our experiments on real data sets show that following such a methodology, interesting outliers can be defined and uncovered flexibly and effectively in large heterogeneous networks.

Systematic discovery of functional modules and context-specific functional annotation of human genome.

MOTIVATION: The rapid accumulation of microarray datasets provides unique opportunities to perform systematic functional characterization of the human genome. We designed a graph-based approach to integrate cross-platform microarray data, and extract recurrent expression patterns. A series of microarray datasets can be modeled as a series of co-expression networks, in which we search for frequently occurring network patterns. The integrative approach provides three major advantages over the commonly used microarray analysis methods: (1) enhance signal to noise separation (2) identify functionally related genes without co-expression and (3) provide a way to predict gene functions in a context-specific way. RESULTS: We integrate 65 human microarray datasets, comprising 1105 experiments and over 11 million expression measurements. We develop a data mining procedure based on frequent itemset mining and biclustering to systematically discover network patterns that recur in at least five datasets. This resulted in 143,401 potential functional modules. Subsequently, we design a network topology statistic based on graph random walk that effectively captures characteristics of a gene's local functional environment. Function annotations based on this statistic are then subject to the assessment using the random forest method, combining six other attributes of the network modules. We assign 1126 functions to 895 genes, 779 known and 116 unknown, with a validation accuracy of 70%. Among our assignments, 20% genes are assigned with multiple functions based on different network environments. AVAILABILITY: http://zhoulab.usc.edu/ContextAnnotation.

A graph-based approach to systematically reconstruct human transcriptional regulatory modules.

MOTIVATION: A major challenge in studying gene regulation is to systematically reconstruct transcription regulatory modules, which are defined as sets of genes that are regulated by a common set of transcription factors. A commonly used approach for transcription module reconstruction is to derive coexpression clusters from a microarray dataset. However, such results often contain false positives because genes from many transcription modules may be simultaneously perturbed upon a given type of conditions. In this study, we propose and validate that genes, which form a coexpression cluster in multiple microarray datasets across diverse conditions, are more likely to form a transcription module. However, identifying genes coexpressed in a subset of many microarray datasets is not a trivial computational problem. RESULTS: We propose a graph-based data-mining approach to efficiently and systematically identify frequent coexpression clusters. Given m microarray datasets, we model each microarray dataset as a coexpression graph, and search for vertex sets which are frequently densely connected across [theta m] datasets (0 < or = theta < or = 1). For this novel graph-mining problem, we designed two techniques to narrow down the search space: (1) partition the input graphs into (overlapping) groups sharing common properties; (2) summarize the vertex neighbor information from the partitioned datasets onto the 'Neighbor Association Summary Graph's for effective mining. We applied our method to 105 human microarray datasets, and identified a large number of potential transcription modules, activated under different subsets of conditions. Validation by ChIP-chip data demonstrated that the likelihood of a coexpression cluster being a transcription module increases significantly with its recurrence. Our method opens a new way to exploit the vast amount of existing microarray data accumulation for gene regulation study. Furthermore, the algorithm is applicable to other biological networks for approximate network module mining. AVAILABILITY: http://zhoulab.usc.edu/NeMo/.

Integrative Array Analyzer: a software package for analysis of cross-platform and cross-species microarray data.

The rapid accumulation of microarray data translates into an urgent need for tools to perform integrative microarray analysis. Integrative Array Analyzer is a comprehensive analysis and visualization software toolkit, which aims to facilitate the reuse of the large amount of cross-platform and cross-species microarray data. It is composed of the data preprocess module, the co-expression analysis module, the differential expression analysis module, the functional and transcriptional annotation module and the graph visualization module.

Mining coherent dense subgraphs across massive biological networks for functional discovery.

MOTIVATION: The rapid accumulation of biological network data translates into an urgent need for computational methods for graph pattern mining. One important problem is to identify recurrent patterns across multiple networks to discover biological modules. However, existing algorithms for frequent pattern mining become very costly in time and space as the pattern sizes and network numbers increase. Currently, no efficient algorithm is available for mining recurrent patterns across large collections of genome-wide networks. RESULTS: We developed a novel algorithm, CODENSE, to efficiently mine frequent coherent dense subgraphs across large numbers of massive graphs. Compared with previous methods, our approach is scalable in the number and size of the input graphs and adjustable in terms of exact or approximate pattern mining. Applying CODENSE to 39 co-expression networks derived from microarray datasets, we discovered a large number of functionally homogeneous clusters and made functional predictions for 169 uncharacterized yeast genes. AVAILABILITY: http://zhoulab.usc.edu/CODENSE/