MUMI: Multitask module identification for biological networks

Research output: Contribution to journalArticlepeer-review

Standard

MUMI : Multitask module identification for biological networks. / Chen, Weiqi; Zhu, Zexuan; He, Shan.

In: IEEE Transactions on Evolutionary Computation, 07.11.2019.

Research output: Contribution to journalArticlepeer-review

Harvard

APA

Vancouver

Author

Bibtex

@article{684c0c94910e4572ac8989576cf81010,
title = "MUMI: Multitask module identification for biological networks",
abstract = "Identifying modules from biological networks is important since modules reveal essential mechanisms and dynamic processes in biological systems. Existing algorithms focus on identifying either active modules or topological modules (communities), which represent dynamic and topological units in the network, respectively. However, high-level biological phenomena, e.g., functions are emergent properties from the interplay between network topology and dynamics. Therefore, to fully explain the mechanisms underlying the high-level biological phenomena, it is important to identify the overlaps between communities and active modules, which indicate the topological units with significant changes of dynamics. However, despite the importance, there are no existing methods to do so. In this paper, we propose MUMI (MUltitask Module Identification) algorithm to detect the overlaps between active modules and communities simultaneously. Experimental results show that our method provides new insights into biological mechanisms by combining information from active modules and communities. By formulating the problem as a multitasking learning problem which searches for these two types of modules simultaneously, the algorithm can exploit their latent complementarities to obtain bettersearch performance in terms of accuracy and convergence. Our MATLAB implementation of MUMI is available at https://github.com/WeiqiChen/Mumi-multitask-module-identification.",
author = "Weiqi Chen and Zexuan Zhu and Shan He",
year = "2019",
month = nov,
day = "7",
doi = "10.1109/TEVC.2019.2952220",
language = "English",
journal = "IEEE Transactions on Evolutionary Computation",
issn = "1089-778X",
publisher = "Institute of Electrical and Electronics Engineers (IEEE)",

}

RIS

TY - JOUR

T1 - MUMI

T2 - Multitask module identification for biological networks

AU - Chen, Weiqi

AU - Zhu, Zexuan

AU - He, Shan

PY - 2019/11/7

Y1 - 2019/11/7

N2 - Identifying modules from biological networks is important since modules reveal essential mechanisms and dynamic processes in biological systems. Existing algorithms focus on identifying either active modules or topological modules (communities), which represent dynamic and topological units in the network, respectively. However, high-level biological phenomena, e.g., functions are emergent properties from the interplay between network topology and dynamics. Therefore, to fully explain the mechanisms underlying the high-level biological phenomena, it is important to identify the overlaps between communities and active modules, which indicate the topological units with significant changes of dynamics. However, despite the importance, there are no existing methods to do so. In this paper, we propose MUMI (MUltitask Module Identification) algorithm to detect the overlaps between active modules and communities simultaneously. Experimental results show that our method provides new insights into biological mechanisms by combining information from active modules and communities. By formulating the problem as a multitasking learning problem which searches for these two types of modules simultaneously, the algorithm can exploit their latent complementarities to obtain bettersearch performance in terms of accuracy and convergence. Our MATLAB implementation of MUMI is available at https://github.com/WeiqiChen/Mumi-multitask-module-identification.

AB - Identifying modules from biological networks is important since modules reveal essential mechanisms and dynamic processes in biological systems. Existing algorithms focus on identifying either active modules or topological modules (communities), which represent dynamic and topological units in the network, respectively. However, high-level biological phenomena, e.g., functions are emergent properties from the interplay between network topology and dynamics. Therefore, to fully explain the mechanisms underlying the high-level biological phenomena, it is important to identify the overlaps between communities and active modules, which indicate the topological units with significant changes of dynamics. However, despite the importance, there are no existing methods to do so. In this paper, we propose MUMI (MUltitask Module Identification) algorithm to detect the overlaps between active modules and communities simultaneously. Experimental results show that our method provides new insights into biological mechanisms by combining information from active modules and communities. By formulating the problem as a multitasking learning problem which searches for these two types of modules simultaneously, the algorithm can exploit their latent complementarities to obtain bettersearch performance in terms of accuracy and convergence. Our MATLAB implementation of MUMI is available at https://github.com/WeiqiChen/Mumi-multitask-module-identification.

U2 - 10.1109/TEVC.2019.2952220

DO - 10.1109/TEVC.2019.2952220

M3 - Article

JO - IEEE Transactions on Evolutionary Computation

JF - IEEE Transactions on Evolutionary Computation

SN - 1089-778X

ER -