Correlation between aggregated molecular cancer subtypes and selected clinical features

Sarcoma (Primary solid tumor)

16 April 2014 | analyses__2014_04_16

Maintainer Information

Citation Information

Maintained by TCGA GDAC Team (Broad Institute/MD Anderson Cancer Center/Harvard Medical School)

Cite as Broad Institute TCGA Genome Data Analysis Center (2014): Correlation between aggregated molecular cancer subtypes and selected clinical features. Broad Institute of MIT and Harvard. doi:10.7908/C13N223G

Overview

Introduction

This pipeline computes the correlation between cancer subtypes identified by different molecular patterns and selected clinical features.

Summary

Testing the association between subtypes identified by 8 different clustering approaches and 3 clinical features across 104 patients, 2 significant findings detected with P value < 0.05 and Q value < 0.25.

3 subtypes identified in current cancer cohort by 'Copy Number Ratio CNMF subtypes'. These subtypes do not correlate to any clinical features.
6 subtypes identified in current cancer cohort by 'METHLYATION CNMF'. These subtypes do not correlate to any clinical features.
CNMF clustering analysis on sequencing-based mRNA expression data identified 3 subtypes that correlate to 'GENDER'.
Consensus hierarchical clustering analysis on sequencing-based mRNA expression data identified 3 subtypes that correlate to 'AGE'.
3 subtypes identified in current cancer cohort by 'MIRSEQ CNMF'. These subtypes do not correlate to any clinical features.
2 subtypes identified in current cancer cohort by 'MIRSEQ CHIERARCHICAL'. These subtypes do not correlate to any clinical features.
3 subtypes identified in current cancer cohort by 'MIRseq Mature CNMF subtypes'. These subtypes do not correlate to any clinical features.
2 subtypes identified in current cancer cohort by 'MIRseq Mature cHierClus subtypes'. These subtypes do not correlate to any clinical features.

Results

Overview of the results

Table 1. Get Full Table Overview of the association between subtypes identified by 8 different clustering approaches and 3 clinical features. Shown in the table are P values (Q values). Thresholded by P value < 0.05 and Q value < 0.25, 2 significant findings detected.


Clinical Features	Time to Death	AGE	GENDER
Statistical Tests	logrank test	t-test	Fisher's exact test
Copy Number Ratio CNMF subtypes	0.594 (1.00)	0.125 (1.00)	0.135 (1.00)
METHLYATION CNMF	0.349 (1.00)	0.17 (1.00)	0.078 (1.00)
RNAseq CNMF subtypes	0.831 (1.00)	0.662 (1.00)	0.00393 (0.0943)
RNAseq cHierClus subtypes	0.66 (1.00)	0.00479 (0.11)	0.0253 (0.532)
MIRSEQ CNMF	0.959 (1.00)	0.0335 (0.671)	0.278 (1.00)
MIRSEQ CHIERARCHICAL	0.917 (1.00)	0.0234 (0.514)	0.233 (1.00)
MIRseq Mature CNMF subtypes	0.539 (1.00)	0.0763 (1.00)	0.387 (1.00)
MIRseq Mature cHierClus subtypes	0.84 (1.00)	0.0477 (0.907)	0.233 (1.00)

Clustering Approach #1: 'Copy Number Ratio CNMF subtypes'

Table S1. Description of clustering approach #1: 'Copy Number Ratio CNMF subtypes'

Cluster Labels	1	2	3
Number of samples	35	28	40

'Copy Number Ratio CNMF subtypes' versus 'Time to Death'

P value = 0.594 (logrank test), Q value = 1

Table S2. Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	103	31	0.1 - 143.4 (18.1)
subtype1	35	12	0.1 - 108.1 (18.1)
subtype2	28	8	0.1 - 81.0 (13.2)
subtype3	40	11	0.1 - 143.4 (22.6)

Figure S1. Get High-res Image Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

'Copy Number Ratio CNMF subtypes' versus 'AGE'

P value = 0.125 (ANOVA), Q value = 1

Table S3. Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	103	62.4 (12.9)
subtype1	35	63.5 (12.5)
subtype2	28	65.5 (12.7)
subtype3	40	59.3 (13.1)

Figure S2. Get High-res Image Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #2: 'AGE'

'Copy Number Ratio CNMF subtypes' versus 'GENDER'

P value = 0.135 (Fisher's exact test), Q value = 1

Table S4. Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	54	49
subtype1	15	20
subtype2	19	9
subtype3	20	20

Figure S3. Get High-res Image Clustering Approach #1: 'Copy Number Ratio CNMF subtypes' versus Clinical Feature #3: 'GENDER'

Clustering Approach #2: 'METHLYATION CNMF'

Table S5. Description of clustering approach #2: 'METHLYATION CNMF'

Cluster Labels	1	2	3	4	5	6	7
Number of samples	24	19	10	9	12	28	2

'METHLYATION CNMF' versus 'Time to Death'

P value = 0.349 (logrank test), Q value = 1

Table S6. Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	102	30	0.1 - 143.4 (18.1)
subtype1	24	7	0.7 - 70.5 (17.0)
subtype2	19	6	0.5 - 108.1 (15.1)
subtype3	10	4	0.1 - 74.7 (7.2)
subtype4	9	2	1.1 - 143.4 (19.3)
subtype5	12	2	0.1 - 116.9 (22.0)
subtype6	28	9	0.5 - 81.0 (25.1)

Figure S4. Get High-res Image Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #1: 'Time to Death'

'METHLYATION CNMF' versus 'AGE'

P value = 0.17 (ANOVA), Q value = 1

Table S7. Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	102	62.3 (13.0)
subtype1	24	64.0 (15.6)
subtype2	19	64.4 (13.2)
subtype3	10	64.9 (10.8)
subtype4	9	64.2 (12.1)
subtype5	12	53.0 (9.7)
subtype6	28	61.7 (11.7)

Figure S5. Get High-res Image Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #2: 'AGE'

'METHLYATION CNMF' versus 'GENDER'

P value = 0.078 (Chi-square test), Q value = 1

Table S8. Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	53	49
subtype1	11	13
subtype2	5	14
subtype3	8	2
subtype4	5	4
subtype5	8	4
subtype6	16	12

Figure S6. Get High-res Image Clustering Approach #2: 'METHLYATION CNMF' versus Clinical Feature #3: 'GENDER'

Clustering Approach #3: 'RNAseq CNMF subtypes'

Table S9. Description of clustering approach #3: 'RNAseq CNMF subtypes'

Cluster Labels	1	2	3
Number of samples	36	21	26

'RNAseq CNMF subtypes' versus 'Time to Death'

P value = 0.831 (logrank test), Q value = 1

Table S10. Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	83	27	0.1 - 143.4 (18.5)
subtype1	36	13	1.7 - 108.1 (18.1)
subtype2	21	6	0.1 - 143.4 (12.8)
subtype3	26	8	0.1 - 81.0 (25.8)

Figure S7. Get High-res Image Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

'RNAseq CNMF subtypes' versus 'AGE'

P value = 0.662 (ANOVA), Q value = 1

Table S11. Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	83	61.9 (12.5)
subtype1	36	63.2 (14.1)
subtype2	21	61.8 (11.2)
subtype3	26	60.3 (11.4)

Figure S8. Get High-res Image Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #2: 'AGE'

'RNAseq CNMF subtypes' versus 'GENDER'

P value = 0.00393 (Fisher's exact test), Q value = 0.094

Table S12. Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	45	38
subtype1	13	23
subtype2	17	4
subtype3	15	11

Figure S9. Get High-res Image Clustering Approach #3: 'RNAseq CNMF subtypes' versus Clinical Feature #3: 'GENDER'

Clustering Approach #4: 'RNAseq cHierClus subtypes'

Table S13. Description of clustering approach #4: 'RNAseq cHierClus subtypes'

Cluster Labels	1	2	3
Number of samples	34	46	3

'RNAseq cHierClus subtypes' versus 'Time to Death'

P value = 0.66 (logrank test), Q value = 1

Table S14. Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	83	27	0.1 - 143.4 (18.5)
subtype1	34	10	0.1 - 116.9 (24.7)
subtype2	46	16	0.1 - 143.4 (14.6)
subtype3	3	1	4.5 - 44.5 (36.4)

Figure S10. Get High-res Image Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

'RNAseq cHierClus subtypes' versus 'AGE'

P value = 0.00479 (ANOVA), Q value = 0.11

Table S15. Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	83	61.9 (12.5)
subtype1	34	58.7 (11.6)
subtype2	46	65.3 (12.2)
subtype3	3	46.3 (6.7)

Figure S11. Get High-res Image Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #2: 'AGE'

'RNAseq cHierClus subtypes' versus 'GENDER'

P value = 0.0253 (Fisher's exact test), Q value = 0.53

Table S16. Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	45	38
subtype1	23	11
subtype2	22	24
subtype3	0	3

Figure S12. Get High-res Image Clustering Approach #4: 'RNAseq cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

Clustering Approach #5: 'MIRSEQ CNMF'

Table S17. Description of clustering approach #5: 'MIRSEQ CNMF'

Cluster Labels	1	2	3
Number of samples	48	39	16

'MIRSEQ CNMF' versus 'Time to Death'

P value = 0.959 (logrank test), Q value = 1

Table S18. Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	103	31	0.1 - 143.4 (18.1)
subtype1	48	14	0.1 - 116.9 (18.1)
subtype2	39	13	0.1 - 143.4 (20.1)
subtype3	16	4	0.7 - 70.5 (16.2)

Figure S13. Get High-res Image Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #1: 'Time to Death'

'MIRSEQ CNMF' versus 'AGE'

P value = 0.0335 (ANOVA), Q value = 0.67

Table S19. Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	103	62.4 (12.9)
subtype1	48	59.8 (12.7)
subtype2	39	62.7 (12.0)
subtype3	16	69.4 (13.8)

Figure S14. Get High-res Image Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #2: 'AGE'

'MIRSEQ CNMF' versus 'GENDER'

P value = 0.278 (Fisher's exact test), Q value = 1

Table S20. Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	54	49
subtype1	29	19
subtype2	17	22
subtype3	8	8

Figure S15. Get High-res Image Clustering Approach #5: 'MIRSEQ CNMF' versus Clinical Feature #3: 'GENDER'

Clustering Approach #6: 'MIRSEQ CHIERARCHICAL'

Table S21. Description of clustering approach #6: 'MIRSEQ CHIERARCHICAL'

Cluster Labels	1	2	3
Number of samples	53	2	48

'MIRSEQ CHIERARCHICAL' versus 'Time to Death'

P value = 0.917 (logrank test), Q value = 1

Table S22. Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	101	30	0.1 - 143.4 (18.1)
subtype1	53	16	0.1 - 143.4 (18.2)
subtype3	48	14	0.1 - 116.9 (18.1)

Figure S16. Get High-res Image Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #1: 'Time to Death'

'MIRSEQ CHIERARCHICAL' versus 'AGE'

P value = 0.0234 (t-test), Q value = 0.51

Table S23. Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	101	62.5 (13.0)
subtype1	53	65.3 (12.6)
subtype3	48	59.5 (12.7)

Figure S17. Get High-res Image Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #2: 'AGE'

'MIRSEQ CHIERARCHICAL' versus 'GENDER'

P value = 0.233 (Fisher's exact test), Q value = 1

Table S24. Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	52	49
subtype1	24	29
subtype3	28	20

Figure S18. Get High-res Image Clustering Approach #6: 'MIRSEQ CHIERARCHICAL' versus Clinical Feature #3: 'GENDER'

Clustering Approach #7: 'MIRseq Mature CNMF subtypes'

Table S25. Description of clustering approach #7: 'MIRseq Mature CNMF subtypes'

Cluster Labels	1	2	3
Number of samples	45	43	15

'MIRseq Mature CNMF subtypes' versus 'Time to Death'

P value = 0.539 (logrank test), Q value = 1

Table S26. Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	103	31	0.1 - 143.4 (18.1)
subtype1	45	13	0.1 - 116.9 (21.4)
subtype2	43	13	0.1 - 143.4 (18.5)
subtype3	15	5	0.7 - 70.5 (15.1)

Figure S19. Get High-res Image Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

'MIRseq Mature CNMF subtypes' versus 'AGE'

P value = 0.0763 (ANOVA), Q value = 1

Table S27. Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	103	62.4 (12.9)
subtype1	45	59.5 (12.3)
subtype2	43	63.6 (12.0)
subtype3	15	67.7 (15.7)

Figure S20. Get High-res Image Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #2: 'AGE'

'MIRseq Mature CNMF subtypes' versus 'GENDER'

P value = 0.387 (Fisher's exact test), Q value = 1

Table S28. Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	54	49
subtype1	27	18
subtype2	20	23
subtype3	7	8

Figure S21. Get High-res Image Clustering Approach #7: 'MIRseq Mature CNMF subtypes' versus Clinical Feature #3: 'GENDER'

Clustering Approach #8: 'MIRseq Mature cHierClus subtypes'

Table S29. Description of clustering approach #8: 'MIRseq Mature cHierClus subtypes'

Cluster Labels	1	2	3
Number of samples	2	48	53

'MIRseq Mature cHierClus subtypes' versus 'Time to Death'

P value = 0.84 (logrank test), Q value = 1

Table S30. Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

	nPatients	nDeath	Duration Range (Median), Month
ALL	101	30	0.1 - 143.4 (18.1)
subtype2	48	14	0.1 - 116.9 (19.7)
subtype3	53	16	0.1 - 143.4 (18.1)

Figure S22. Get High-res Image Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

'MIRseq Mature cHierClus subtypes' versus 'AGE'

P value = 0.0477 (t-test), Q value = 0.91

Table S31. Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #2: 'AGE'

	nPatients	Mean (Std.Dev)
ALL	101	62.5 (13.0)
subtype2	48	59.9 (13.0)
subtype3	53	65.0 (12.5)

Figure S23. Get High-res Image Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #2: 'AGE'

'MIRseq Mature cHierClus subtypes' versus 'GENDER'

P value = 0.233 (Fisher's exact test), Q value = 1

Table S32. Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

nPatients	FEMALE	MALE
ALL	52	49
subtype2	28	20
subtype3	24	29

Figure S24. Get High-res Image Clustering Approach #8: 'MIRseq Mature cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

Methods & Data

Input

Cluster data file = SARC-TP.mergedcluster.txt
Clinical data file = SARC-TP.merged_data.txt
Number of patients = 104
Number of clustering approaches = 8
Number of selected clinical features = 3
Exclude small clusters that include fewer than K patients, K = 3

Clustering approaches

CNMF clustering

consensus non-negative matrix factorization clustering approach (Brunet et al. 2004)

Consensus hierarchical clustering

Resampling-based clustering method (Monti et al. 2003)

Survival analysis

For survival clinical features, the Kaplan-Meier survival curves of tumors with and without gene mutations were plotted and the statistical significance P values were estimated by logrank test (Bland and Altman 2004) using the 'survdiff' function in R

ANOVA analysis

For continuous numerical clinical features, one-way analysis of variance (Howell 2002) was applied to compare the clinical values between tumor subtypes using 'anova' function in R

Fisher's exact test

For binary clinical features, two-tailed Fisher's exact tests (Fisher 1922) were used to estimate the P values using the 'fisher.test' function in R

Chi-square test

For multi-class clinical features (nominal or ordinal), Chi-square tests (Greenwood and Nikulin 1996) were used to estimate the P values using the 'chisq.test' function in R

Student's t-test analysis

For continuous numerical clinical features, two-tailed Student's t test with unequal variance (Lehmann and Romano 2005) was applied to compare the clinical values between two tumor subtypes using 't.test' function in R

Q value calculation

For multiple hypothesis correction, Q value is the False Discovery Rate (FDR) analogue of the P value (Benjamini and Hochberg 1995), defined as the minimum FDR at which the test may be called significant. We used the 'Benjamini and Hochberg' method of 'p.adjust' function in R to convert P values into Q values.

Download Results

In addition to the links below, the full results of the analysis summarized in this report can also be downloaded programmatically using firehose_get, or interactively from either the Broad GDAC website or TCGA Data Coordination Center Portal.

References

[1] Brunet et al., Metagenes and molecular pattern discovery using matrix factorization, PNAS 101(12):4164-9 (2004)

[2] Monti et al., Consensus Clustering: A Resampling-Based Method for Class Discovery and Visualization of Gene Expression Microarray Data, Machine Learning 52(1):91-118 (2003)

[3] Bland and Altman, Statistics notes: The logrank test, BMJ 328(7447):1073 (2004)

[4] Howell, D, Statistical Methods for Psychology. (5th ed.), Duxbury Press:324-5 (2002)

[5] Fisher, R.A., On the interpretation of chi-square from contingency tables, and the calculation of P, Journal of the Royal Statistical Society 85(1):87-94 (1922)

[6] Greenwood and Nikulin, A guide to chi-squared testing, Wiley, New York. ISBN 047155779X (1996)

[7] Lehmann and Romano, Testing Statistical Hypotheses (3E ed.), New York: Springer. ISBN 0387988645 (2005)

[8] Benjamini and Hochberg, Controlling the false discovery rate: a practical and powerful approach to multiple testing, Journal of the Royal Statistical Society Series B 59:289-300 (1995)

Made with Nozzle