| Cancer is driven by somatic alterations,which accumulate in a person’s life.In cancer genome,an important component of carcinogenic mechanism discovering and clinical treatment option findings is to distinguish driver genes from passenger genes.There is mutual exclusivity between part of the cancer genes.Two hypotheses of mutual exclusivity have been proposed.One of them is that the mutually exclusive genes are functionally linked in a pathway,and multiple mutation events are functional redundancy.The other hypothesis is that cooccurrence of some alterations is detrimental to cancer cells,leading to the synthetic lethal.Thus,the phenomenon of mutual exclusivity among genes can be used as an indicator of driver genes.There are two main categories of mutually exclusive genes identification procedures.The first one is the de novo methods,and the other one is algorithms with biological knowledge.The de novo findings do not constrained by the preknowledge,and new driver genes which have not been discussed by other methods are more likely to be discovered by this kind of methods.Three new mutually exclusive gene identification algorithms are proposed,i.e.,FSME,MeQP,and SME.FSME is based on an independence test,and MeQP uses the quadratic programming,both of which are de novo methods to find genes with mutual exclusivity.SME deals with the problem where cancer subtypes are involved in the identification procedure and identifies the genes with mutual exclusivity under the stratification of cancer subtypes.De novo methods only require mutation matrices as the input,which record every gene’s mutation status on each patient.Ideally,there is one and only one mutated gene in a mutually exclusive gene set for every patient.But in reality,not all patients have the phenomenon of mutual exclusivity.The percentage of the patients with mutual exclusivity is called coverage.FSME uses distance measures to identify candidate mutually exclusive gene pairs,and expands them with forward stagewise selection procedure.An independence test is applied by FSME to distinguish the expanded gene sets with mutual exclusivity.FSME has excellent performances in identifying mutually exclusive genes on simulation data sets,and it is also efficient.The simulation data sets encompass several different mutual exclusivity scenarios:one mutually exclusive gene set,two independent mutually exclusive gene sets,and two mutually exclusive gene sets with overlap gene.The precision of FSME is high on each of these scenarios,and the recall increases with the coverage rate.Moreover,the false positive rates of FSME is stable over diverse coverage rates.FSME is also applied to real cancer data sets,where it can identify gene sets with biological explanations.Robust analysis shows that the number of candidate genes has little influence on the results of FSME.MeQP is proposed to further improve the computation efficiency.Simulation results indicate that the F-measure of MeQP is comparable to that of FSME.Given the fact that MeQP needs less time to produce a result,this means that MeQP is preferred in the situation where the size of candidate genes or the sample size is large.Some types of cancer are complex and heterogeneous in their own categories.Cancer subtypes are used to classify patients and build corresponding subtypespecific models.With the information of subtypes,the genes with mutual exclusivity are more likely to be on the same pathway,instead of having mutual exclusivity due to gene alterations accumulated in specific subtypes respectively.SME is proposed to find mutually exclusive genes which have this tendency in all subtypes,and filter out the ones which only mutate in some specific subtypes.Compared with other strategies which also deals with the subtype information,SME has high precision over different coverage rates.Furthermore,on mutual exclusivity overall but not within subtypes data sets the performance of SME is satisfactory because it controls false positives well under all coverage rates.The mutually exclusive genes identification procedure is applied on breast cancer real data set.Gene DST is grouped with tumor suppressor gene TP53 in the analysis.While gene DST encodes protein,there is little research on its influence on breast cancer.Further analysis on multiple data sets indicates that DST is a promising driver gene for breast cancer.Particularly,DST has more impact on the patients with the luminal-B subtype. |
