Survival scan

A tool to find genes which might predict differential survival when patients from a group are divided in up-/down-regulated subgroups.

Assume your hybridisations are from a set of patients with defined survival and censoring state.
Survival scan calculates Mean and Standard Deviation of regulation from all selected patients.

Next the group is sub-divided in two arms:

• Up: Patients with significant up regulation compared to whole consortium

• Down: Patients with significant down regulation compared to whole consortium

for all three subgroups (up/down/all) Kaplan-Meier curves and Median survival  are computed.

Now you can filter those genes with predict differential survival in groupings

up <=> all
down <=> all
up <=> down

Click Survival button and Select Survival filter from drop-down menu:

The Group selection dialog pops up.
Here you can define two groups:

• Group 1 : The diagnostic group, i.e. those hybridisations (patients) which will be analysed and assigned to the two sub-classes

• Group 2 : Additional hybridisations used to calculate reference survival data

On the parameters tab-sheet define required values:

• Survival time row, Censoring row like for calculation of Kaplan-Meier curves

• Threshold: Algorithm how average expression for a single gene is calculates: Mean=arithmetic mean, Median= Median mean

• Min.Group size: Minimum number of hybs (patients in the two sub-classes (up- / down-regulated)

• Confidence interval: Number of standard deviations for thresholding

Algorithm:

For each single gene the following steps are performed:

1. From all members of Group-1 calculate:

• Average (either Mean or Median)

• Standard Deviation

2. Assign members from group 1 to:

• Class1: Up-regulated = Respective Hyb (patient) has expression value >= Average + (Conf.Interval * Standard-Deviation)

• Class2: Down-regulated = Respective Hyb (patient) has expression value <= Average - (Conf.Interval * Standard-Deviation)

• No-class: Not regulated = all other

3. Filter on Class-size: If (member-number from Class1 < Min.Group-Size) or (member-number from Class2 <Min.Group-Size) => ignore this gene

4. Compute median survival for members from Class1 (=MS1), Class2 (=MS2), Group1+Group2 (=MSR)

5. Compute survival-gain / -loss (SGL) factor:

• SGL1 = MS1 / MSR, if SGL1>1 => SGL1 = 1/SGL1

• SGL2 = MS2 / MSR, if SGL2>1 => SGL2 = 1/SGL2

• SGL12 = MS1 / MS2, if SGL12>1 => SGL12 = 1/SGL12

These values are later on shown in the p-graph. As smaller the SGxx values, as larger the gain or loss of survival.

Analyse your results:

Open the analysis in the analysis tree.

On the p-graph adjust the gain / loss factor (i.e. adjust the p-value slider).
For technical reasons the the p-graph was "misused and not correctly adapted. Instead o the normal statistical p-value a gain loss factor is displayed (as computed by SUMO, see above)
In the example: p=0.4 => onyl genes with loss or gain of survival >= 60% compared to the reference group will be selected.

Select the SGL-class by selecting the respective field in the result table: down<=>ref, up<=>ref, down<=>up (here: down-ref)

Click Kaplan-Meier survival curves in the analysis-tree to view Kaplan-Meier curves from the filtered gene for the three classes:

Use the navigation buttons (blue arrow buttons) to cycle through the filtered genes.

Click Population map to visualise which hybs (patients) are members of the Down-/Up-regulated class for the individual filtered gene.
The population may be clustered like a regular heat map:

Colours indicate membership of a single hyb (patient) in a particular gene to the two (up- / down- regulated versus average) class.
Also all other functionality available for handling and analysing of heat maps can be used.

Survival Scan 2

Applied to 3 groups.
Find genes differentially regulated in patient group 2 <=> 3.
Try to find genes in group1 which might predict differential survival when patients from group1 are divided in Group2/3 similar patients.

Assume your hybridisations are from a set of patients with defined survival and censoring state.
Survival scan calculates Mean and Standard Deviation of regulation from all selected patients.

Two patient groups can be defined with the aim to find genes which are differentially regulated under patient parameters which were used to sub-divide the two arms:
E.g. you have hybridisation from liver-tumour samples. Your patient annotations contain information about Grading (=differntiation of tumour tissue: 1=differntiated=looks like normal liver ,...,3,4=undifferentiated no similarity to normal liver)

A 2-class t-test without multiple testing correction is used to find differentially expressed genes at the defined critical p-value (G1G3-alpha) between Grade1 / Grade34 patients.

Next, the patients in group 2 (e.g. in our example patients with Grade 2) are divided in those being more similar to the Grade1 / Grade34 patients.
This is done by computing Mean and SDev between G1 / G34. Now, all patients
    >  M+Confidence range*SDev => Subgroup1
    <  M-Confidence range*SDev => Subgroup2

As above Kaplan-Meier, median surviaval and Log-Rank test are calculated for Subgroup1/2 and Group1.

---

Survival filter 4

Instead of filtering genes on their statistically significant differntial regulation between two groups, we use the correlation between differential regulataion and patient survival as validator.

Select "Survival filter 4":


The selection dialog open up.
In the usual way, select all samples (patients) which shall be used for the analysis.

Go to the parameters tab-sheet and define:

Survival time row	Sample annotation row containing survival data Days, weeks, month, years wahtever. The row shold only contain NUMBERS (5 weeks is not a number and will be interpreted as 0)
Censored time row	Sample annotation row containing survival data for censored samples. Instead you may use the censoring row.
Censoring row	Sample annotation row containing consoring nformation, i.e. defines whether the individual died at that time point (Survival time point) or was lost from the study without further knowledge about survival. Use "yes", "ja", "c", "censored","1" to define a sampel as censored, anything else is interpreted as not censored.
Threshold	Define how to group the samples: Mean: compute aritmetic mean and SDev. use Mean -/+ConfidenceInterval*SDev as threshold for Down/Up regulated samples. Median: Same as above, but use Median instead of artihmetic mean. Geometric: Same as above, but use Geometric mean. Harmonic: use Harmonic mean. RMS: Root-Mean-Square mean. Quartile: Sort samples by expression, 25% lowest expressed => Class1 25% highest expressed => Class2 Percentile=xx: Generalization of quartile: Sort samples by expression xx% lowest expressed => Class1 xx% highest expressed => Class2 e.g. "percentile=25" would give exactly the same as Quartile Data range 1% - 50%, default=25% Absolute=xx: Similar to Percentile, but take take absolute number of samples. e.g. "absolute=17" would assig the 17 lowest/highest regulated genes inot calsses 1/2 Data range: 1 - (Number of samples /2), default=10
G2/G3.alpha	not used
Min group size	The minimal number of samples per group (down/up regulated samples in a particular gene)

Click RUN button


SUMO will now compute for each individual gene:

• Average regulation / Dev from regulation
• sort the data
• Assign samples to two groups:
  > down - all samples where regulation < Threshold
  > up - all samples where regulation > Threshold
• Ignore genes where either Down/Up group contains too few members
• Compute 2-class t-test between Down/Up group
• Compute Median Survival for Down/Up/Combined group
• Compute pairwise survival gain/loss between Down/Up/Combined groups
• Compute pairwise survival Log Rank test p-value between Down/Up/Combined groups

Use p-graph to filter genes on survival gain/loss or log rank test p-value:


The blue lines show distribution of log2 from abolsute Median-Survival Gain/Loss scaled by factror of 0.1.
I.e. -10 in the graph corresponds to log2=1 => 2 fold difference in Median Survival between the two calsses.

The green lines visualize distribution of p-values from log rank test.

For each set values are shown coparing:

• Down-regulated vs. all
• Up-regulated vs. all
• Down vs. Up -regulated

In the table, click the corresponding cells, to lateron preview the respective

• Kaplan-Meier survival curves
,
• Population map
• Selected genes

Kaplan-Meier survival curves:

Use the navigation buttons (blue arrow buttons) to cycle through the filtered genes.

Population map: visualize which hybs (patients) are members of the Down-/Up-regulated class for the individual filtered gene.
The population may be clustered like a regular heat map.

Colours indicate membership of a single hyb (patient) in a particular gene to the two (up- / down- regulated versus average) class.
Also all other functionality available for handling and analysing of heat maps can be used.

Selected genes: a table summarizing the computationa results for the selected genes:

File | Save to save the table.

The individual columns contain:

Name	Value (exsample)	Meaning
ReporterID	ebv-miR-BART11-5p	Name of gene/....
Source line #	66	Gene-ID in original data matrix
# analysed	495	Number of samples analyzed; shoud be identical for all genes
Down threshold	-0.008	Threshold for down-regulated samples
# down	29	Number of down-regulated samples
% down	0.06	Number of down-regulated samples in %
Members	12:2361:0,20:430:0,22:326:0,...	A data triplet for each sample within the group: Sample-ID:Survival-time:Censor-State e.g. "12:2361:0" - Sample 12, SUrvival time=2361 (days), not cnesored (0)
Up threshold	0.010	Threshold for up-regulated samples
# up	38	Number of up-regulated samples
% up	0.08	Number of up-regulated samples in %
Members	1:358:0,5:705:0,6:322:0,39:167:0,...	Sample-IDs,Survival,Censorstate for all members
G1-Mean	-0.011	Mean (arithmetic) regulation in Down-regulated samples (Group1)
G1-SDev	0.002	Standard deviation from regulation in Group1
G2-Mean	0.019	Mean (arithmetic) regulation in Up-regulated samples (Group2)
G2-SDev	0.016	Standard deviation from regulation in Group2
G1-G2	-0.030	Differential regulation (Mean-G1 - Mean-G2)
t-value	-11.538	t-value (Students-t-test) with Welsh approximation)
p-value	1.91E-014	p-value from t-distribution with Sattertwaithe approximation)
#-reference	383	Number of samples used to compute reference Kaplan Meier Curve (KM) Sample numbers used for KM-data may differ from groupsize. Samples, for which there were NO survaval data in sample annotation, can not be used for survival data (obvious)
Members1	12 20 22 ...	List of sample-ID for G1; may be used to identidfy samples in Heatmap/ Group Selector
Members2	1 5 6 39 ...	List of sample-ID for G2
Down-Median survival	482.000	Median survival in G1
Up-Median survival	350.000	Median survival in G2
Reference-Median survival	377.000	Median survival in ALL selectged samples (Ref)
Down/Reference log2-survival-gain	0.782	Compute: Ratio=MediaSurvial-G1 / Media_Survival-Ref This cell contains = absolute (log2 (Ratio)) Cell value=1 => The median-survival betwen the groups differs by factor 2x
Up/Reference log2-survival-gain	0.928
Down/Up log2-survival-gain	0.726
Down-Reference Log rank test	1.037	Log Rank Test (LRT) between G1<=>Ref (Down-Regulated<=>Ref)
Up-Reference Log rank test	4.254	Rank Test between G2<=>Ref (Up-Regulated<=>Ref)
Down-Up Log rank test	5.952	Rank Test between G1<=>G2 (Down<=>up-Regulated)
Down-Reference p-LRT	1.00E+000	p-value from LRT G1<=>Ref
Up-Reference p-LRT	1.00E+000	p-value from LRT G2<=>Ref
Down-Up p-LRT	1.47E-002	p-value from LRT G1<=>G2

---

---