A model and algorithm for identifying driver pathways based on weighted non-binary mutation matrix

Abstract

It is generally acknowledged that driver pathway plays a decisive role in the occurrence and progress of tumors, and the identification of driver pathways has become imperative for precision medicine or personalized medicine. Due to the inevitable sequencing error, the noise contained in single omics cancer data usually plays a negative effect on identification. It is a feasible approach to take advantage of multi-omics cancer data rather than a single one now that large amounts of multi-omics cancer data have become available. The identification of driver pathways by integrating multi-omics cancer data has attracted attention of researchers in bioinformatics recently. In this paper, a weighted non-binary mutation matrix is constructed by integrating copy number variations, somatic mutations and gene expressions. Based on the weighted non-binary mutation matrix, a new identification model is proposed through defining new measurements of coverage and exclusivity. Then, a cooperative coevolutionary algorithm CGA-MWS is put forward for solving the presented model. Both real cancer data and simulated one were used to conduct comparisons among methods Dendrix, GA, iMCMC, MOGA, PGA-MWS and CGA-MWS. Compared with the pathways identified by the other five methods, more genes, belonging to the pathway identified by the CGA-MWS method, are enriched in a known signaling pathway in most cases. Simultaneously, the high efficiency of method CGA-MWS makes it practical in realistic applications. All of which have been verified through a number of experiments.

Identifying driver pathways based on a parameter-free model and a partheno-genetic algorithm

“…We begin by testing the models which are based on the proposed coverage and mutual exclusivity, and comparing them with the famous one proposed by Vandin et al [5], which has also been used in such methods as Dendrix [5], GA [7], and MOGA [9]. Then the identification performance of method CPGA-SMCMN was compared with six state-of-the-art methods, i.e., Dendrix [5], CGA-MWS [6], GA [7], iMCMC [8], MOGA [9] and PGA-MWS [11]. The experimental comparisons were implemented on a Lenovo PC with Intel(R) Core(TM) i7-7700 3.60GHz CPU and 24GB RAM.…”

“…In this section, the identification performance was compared among methods Dendrix [5], GA [7], iMCMC [8], MOGA [9], PGA-MWS [11], CGA-MWS [6] and CPGA-SMCMN. In addition, the performance of algorithm CPGA for solving the classical MWSM model was also tested and presented.…”

“…In Figure 1, M E(B)= 16 27 , and M E(C)= 83 120 . The obvious choice is matrix C for its larger M E(M ) using Formula (6).…”

“…Since it is time-consuming as well as expensive to identify through biological experiments in the lab, it is a very economic way to detect driver pathways (driver gene sets) by applying computational approaches on the abundant accumulated multi-omics data. This has received widely concern in bioinformatics [4,5,6].…”

“…In 2019, Wu et al [11] improved the maximum weight submatrix model and proposed method PGA-MWS for solving this problem. In 2021, Wu et al [6] introduced a weighted non-binary mutation matrix. They formulated a new maximum weight submatrix model by redefining coverage and mutual exclusivity, and devised a cooperative co-evolution algorithm CGA-MWS for solving this model.…”