MUMI: Multitask Module Identification for Biological Networks

Weiqi Chen,Zexuan Zhu,Shan He

DOI: https://doi.org/10.1109/tevc.2019.2952220

IF: 16.497

2020-08-01

IEEE Transactions on Evolutionary Computation

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 article, we propose the multitask module identification (MUMI) algorithm to detect the overlaps between active modules and communities simultaneously. The 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 better search performance in terms of accuracy and convergence. Our MATLAB implementation of MUMI is available at https://github.com/WeiqiChen/Mumi-multitask-module-identification.

computer science, artificial intelligence, theory & methods

What problem does this paper attempt to address?

The problem this paper attempts to address is how to simultaneously identify the overlapping parts between communities (i.e., topological modules) and active modules in biological networks. Existing algorithms typically focus only on identifying active modules or topological modules, but these methods cannot comprehensively explain advanced biological phenomena (such as functions) because these phenomena are generated by network topology and dynamic interactions. Therefore, by detecting the overlapping parts between communities and active modules, the mechanisms behind these advanced biological phenomena can be revealed.

Specifically, the paper proposes the following hypotheses:
1. By identifying the overlapping parts between communities and active modules, the mechanisms of advanced biological phenomena that cannot be obtained by identifying these modules alone can be revealed.
2. Simultaneously searching for communities and active modules can improve the accuracy and convergence of module identification, as network topology and dynamics complement each other to jointly produce advanced biological phenomena.

To verify these hypotheses, the authors propose a new multi-task module identification algorithm based on multi-factor evolution (MUMI). This algorithm can simultaneously identify communities and active modules in biological networks and utilize their potential complementarity to improve search performance. Experimental results show that the MUMI algorithm can provide new insights into biological mechanisms and perform excellently in terms of accuracy and convergence.