Guyu Hu | Searcharxiv

arXiv2016

MODA: MOdule Differential Analysis for weighted gene co-expression network

Gene co-expression network differential analysis is designed to help biologists understand gene expression patterns under different condition. By comparing different gene co-expression networks we may find conserved part as well as condition specific set of genes. Taking the network as a collection as modules, we use a sample-saving method to construct condition-specific gene co-expression network, and identify differentially expressed subnetworks as conserved or condition specific modules which may be associated with biological processes. We have implemented the method as an R package which establishes a pipeline from expression profile to biological explanations. The usefulness of the method is also demonstrated by synthetic data as well as Daphnia magna gene expression data under different environmental stresses.

q-bio.QM↗

arXiv2014

Multi-task Feature Selection based Anomaly Detection

Network anomaly detection is still a vibrant research area. As the fast growth of network bandwidth and the tremendous traffic on the network, there arises an extremely challengeable question: How to efficiently and accurately detect the anomaly on multiple traffic? In multi-task learning, the traffic consisting of flows at different time periods is considered as a task. Multiple tasks at different time periods performed simultaneously to detect anomalies. In this paper, we apply the multi-task feature selection in network anomaly detection area which provides a powerful method to gather information from multiple traffic and detect anomalies on it simultaneously. In particular, the multi-task feature selection includes the well-known l1-norm based feature selection as a special case given only one task. Moreover, we show that the multi-task feature selection is more accurate by utilizing more information simultaneously than the l1-norm based method. At the evaluation stage, we preprocess the raw data trace from trans-Pacific backbone link between Japan and the United States, label with anomaly communities, and generate a 248-feature dataset. We show empirically that the multi-task feature selection outperforms independent l1-norm based feature selection on real traffic dataset.

stat.ML↗