Important update (January 20th, 2011): the data below have been corrected for the BCR batch which is not necessarily the processing batch. The dataset needs to be reanalyzed.
Correlation between BCR batch and the processing batch for 27k arrays (January 20, 2012)
Batch 47,1,A00Y
Batch 56,2,A032
Batch 61,no data,
Batch 72,no data,
Batch 74,no data,
Batch 80,no data,
Batch 85,3,A112
Batch 93,4,A12E
Batch 96,no data,
Batch 103,no data,
Batch 109,no data,
Batch 117,no data,
Batch 120,no data,
Batch 124,no data,
Batch 136,no data,
Batch 142,no data,
Batch 147,no data,
Batch 155,no data,
Batch 167,no data,
Batch 177,no data,
Batch 185,no data,
Analysis of batch vs clinical traits
Number of clinical traits: 84
Number of batches based on tumor DNA methylation data (samples retrieved according to this pattern: "TCGA-......-0....D-....-05"): 24
Correlation between center and batches ('two'=center (second field in the patient barcode)):
Significant batch-clinical traits correlations (the entire list can be found here):
"BRCA_clinical_traits","DataType","NumberOfNAs","Test","Pvalue"
"tissue_prospective_collection_indicator","factor",35,"Pearson's Chi-squared test",4.47E-62
"tissue_retrospective_collection_indicator","factor",35,"Pearson's Chi-squared test",4.47E-62
"year_of_initial_pathologic_diagnosis","integer",34,"Kruskal-Wallis rank sum test",3.15E-32
"breast_carcinoma_first_surgical_procedure_name","factor",54,"Pearson's Chi-squared test",5.45E-32
"days_to_last_followup","integer",73,"Kruskal-Wallis rank sum test",3.07E-31
"days_to_form_completion","integer",34,"Kruskal-Wallis rank sum test",5.70E-31
"first_pathologic_diagnosis_biospecimen_acquisition_method_type","factor",123,"Pearson's Chi-squared test",3.39E-28
"breast_tumor_clinical_m_stage","factor",35,"Pearson's Chi-squared test",1.06E-22
"axillary_lymph_node_stage_method_type","factor",223,"Pearson's Chi-squared test",9.33E-19
"breast_tumor_pathologic_n_stage","factor",34,"Pearson's Chi-squared test",2.19E-17
"lab_proc_her2_neu_immunohistochemistry_receptor_status","factor",41,"Pearson's Chi-squared test",6.22E-16
"breast_carcinoma_estrogen_receptor_status","factor",34,"Pearson's Chi-squared test",1.85E-13
"breast_carcinoma_progesterone_receptor_status","factor",34,"Pearson's Chi-squared test",8.87E-13
"vital_status","factor",34,"Pearson's Chi-squared test",2.38E-09
"anatomic_site_location_descriptor","factor",119,"Pearson's Chi-squared test",1.03E-07
"age_at_initial_pathologic_diagnosis","integer",34,"Kruskal-Wallis rank sum test",5.87E-06
"days_to_birth","integer",34,"Kruskal-Wallis rank sum test",6.68E-06
"lab_procedure_her2_neu_in_situ_hybrid_outcome_type","factor",194,"Pearson's Chi-squared test",3.18E-05
"person_menopause_status","factor",161,"Pearson's Chi-squared test",5.70E-05
"breast_tumor_pathologic_grouping_stage","factor",40,"Pearson's Chi-squared test",7.40E-05
"her2_immunohistochemistry_level_result","factor",351,"Pearson's Chi-squared test",1.72E-04
"breast_tumor_pathologic_t_stage","factor",34,"Pearson's Chi-squared test",2.82E-04
"pos_finding_lymph_node_hematoxylin_and_eosin_staining_microscopy_count","integer",177,"Kruskal-Wallis rank sum test",6.49E-04
"cytokeratin_immunohistochemistry_staining_method_micrometastasis_indicator","factor",324,"Pearson's Chi-squared test",8.61E-04
"person_neoplasm_cancer_status","factor",284,"Pearson's Chi-squared test",7.95E-03
"breast_cancer_optical_measurement_histologic_type","factor",34,"Pearson's Chi-squared test",1.47E-02
"disease_surgical_margin_status","factor",82,"Pearson's Chi-squared test",3.70E-02
Correlation with survival
Relevant clinical traits: days to the last follow-up (27), vital status (83), days to death (24), days to last know alive (28), summaries:
It seems that similarly to the colon cancer combined datasets days to last known alive is similar to the days to the last follow-up, however days to the last follow up contains more information (fewer NAs), use it for construction of the survival object. No patients missed information for both days to the last follow up and days to death. The survival object was created in the same way as for the analyses of other TCGA cancer datasets. Info is available (here and here)
Kaplan Meier curve and survival plots break down by batch:
Here is the summary of the survival vs batch:
It seems that there are a lot of errors, I wonder why. I also don't understand where those observations come from that are deleted due to missingness. Need to ask someone to help clarify this output. Update (January 5, 2012): there are NAs for some batches because I had factor levels left in the batch vector but no data for those levels. Fixed the problem with that. "Deleted due to missingness" also fixed as I figured out how that I need to be more careful about using 'match' for subsetting.
DNA methylation data
December 21st, 2011: 27k and 450k arrays are available. Downloaded Level 1 450k data. It seems that they started splitting green and red probes into 2 separate files and they also provide now the Illumina's idat files which are the bead level data (not tab delimited files). I need to find a way to process them, it seems that Bioconductor beadarray package can be used to read these files and do some bead level normalization (summarization too?). The Level2 data contains already summarized and normalized data (tab delimited files with CpG ID, value for methylated and value for unmethylated probes), however it is available only for 91 patients. Also tried to download 27k arrays available for breast cancer, however the data is available for ~26 patients (they stopped running those arrays?). I guess I need to figure out how to process Level 1 data.