GSEA result with NaN NES and NOM p-value

[Yao Wei Lu]

Hi, gsea-help,

I encountered a problem while running my data-set using c5.all.v6.2.symbols.gmt. The result returned for the enrichment on my control group looks strange, as the gene set towards the top has "NaN" as the NES, by looking at the plot they seem to be enrich. I wonder if there's a way to fix this problem, at least to have the correct NES for those gene-sets to be display. Below are some screen shot for the result and the parameter of the run.

producer_class	xtools.gsea.Gsea
producer_timestamp	1532207211321
param	collapse	true

param	plot_top_x	20
param	norm	meandiv
param	save_rnd_lists	false
param	median	false
param	num	100
param	scoring_scheme	weighted
param	make_sets	true
param	mode	Max_probe
param	gmx	gseaftp.broadinstitute.org://pub/gsea/gene_sets_final/c5.all.v6.2.symbols.gmt
param	gui	false
param	chip	gseaftp.broadinstitute.org://pub/gsea/annotations/ENSEMBL_mouse_gene.chip
param	metric	Signal2Noise
param	rpt_label	CIP_KO_mdx_vs_control_mdx_c5_Ontology_v6.2
param	help	false
param	order	descending

param	create_svgs	false
param	permute	gene_set
param	rnd_type	no_balance
param	set_min	15
param	include_only_symbols	true
param	sort	real
param	create_gcts	false
param	rnd_seed	timestamp
param	nperm	1000
param	zip_report	false
param	set_max	500

[Yao Wei Lu]

Software version: GSEA v3.0 Build 0160

[ptamayo]

Yao,

  This is a rare case but it can happen with very large gene sets and a skew distribution of gene scores. The problem is that because of those two conditions the permutation test is only populating one side (positive or negative) of the null distribution. When it divides the ES by the mean of the corresponding side of the null distribution to compute the NES is gets a  division by zero.

  I'll suggest that you try using the phenotype instead of gene permutations and increase the number of permutations to 10,000 or higher.

 Let us know if this solves the problem.

Best,

--Pablo Tamayo

[Yao Wei Lu]

Thanks for your help Pablo. Increase the number of permutations to 40,000 solves the problem.

Best,

Yao Wei

[rba...@gmail.com]

Hi,

I am having NaN for NES values as well. I have only 3 samples per group so I believe this means that I should keep the permutation type as gene_set. However, when I try to increase the number of permutations to 10,000, I got the below error message for not enough Java heap space. Any idea how I can overcome this?

Thanks,

Robin

<Error Details>

---- Full Error Message ----

Java heap space

---- Stack Trace ----

# of exceptions: 1

------Java heap space------

java.lang.OutOfMemoryError: Java heap space

at edu.mit.broad.genome.math.Vector.<init>(Vector.java:80)

at edu.mit.broad.genome.math.Vector.<init>(Vector.java:214)

at edu.mit.broad.genome.math.Vector.<init>(Vector.java:199)

at edu.mit.broad.genome.alg.gsea.KSCore.calculateKSScore_all_modes(KSCore.java:262)

at edu.mit.broad.genome.alg.gsea.KSCore.calculateKSScore(KSCore.java:30)

at edu.mit.broad.genome.alg.gsea.KSTests.shuffleGeneSet_precannedRankedList(KSTests.java:290)

at edu.mit.broad.genome.alg.gsea.KSTests.shuffleGeneSet(KSTests.java:341)

at edu.mit.broad.genome.alg.gsea.KSTests.executeGsea(KSTests.java:105)

at edu.mit.broad.genome.alg.gsea.KSTests.executeGsea(KSTests.java:76)

at xtools.gsea.AbstractGsea2Tool.execute_one(AbstractGsea2Tool.java:121)

at xtools.gsea.AbstractGsea2Tool.execute_one_with_reporting(AbstractGsea2Tool.java:159)

at xtools.gsea.Gsea.execute(Gsea.java:116)

at edu.mit.broad.xbench.tui.TaskManager$ToolRunnable.run(TaskManager.java:436)

at java.lang.Thread.run(Thread.java:748)

[Anthony Castanza]

The GSEA website offers several options to launch GSEA with increased memory allocations:

 http://software.broadinstitute.org/gsea/downloads.jsp (see the "Launch With" dropdown menu)

Alternatively, you can override the Java defaults to run GSEA desktop with increased heap space by launching the GSEA Jar file through the command line with the -Xmx####m flag. Replace the #### with the number of megabytes available for your system (I believe that the java default is 1024).

-Anthony

Anthony S. Castanza, PhD

Curator, MSigDB

Mesirov Lab, Department of Medicine

University of California, San Diego