Breast cancer with bone metastasis

[ddolbae01]

Study title: Identification of causal interaction of osteoblast related genes in bone metastasis of breast cancer.

Backgrounds
1. Seventy percent of cancer patients have detectable metastases when they receive a diagnosis and 90% of cancer deaths result from metastases.
2. Approximately half of these newly diagnosed cases will involve the skeleton.
3. Although bone metastases can occur in conjunction with any solid tumor, they are most common in breast, prostate, and lung cancer patients.
4. There is increasing evidence that osteoblasts are important components of the bone metastatic niche, but their precise contribution in supporting tumour cell engraftment, dormancy and survival remains to be defined.

Methods
1. Obtaining three groups of gene data from human sample: bone metastasis of breast cancer (BM group), breast cancer without bone metastasis (B group) and normal osteoblast (Ob group) (obtained from GEO)
2. Finding commonly expressed genes : the genes should be commonly expressed gene between BM and Obs groups. But the gene should not expressed in B group.
3. Making causal interaction between genes – Using Bayesian network
4. Consideration of future study: verify the genes in breast cancer cell lines that are obtained from breast cancer-bone metastasis mouse model.

present processing
- I attached the excel file that has list of GSE and GSM of three groups in first sheet (obtained from GEO searching).

lis of GSE and GSM.xlsx

(20.25 KiB) Downloaded 143 times

[lsand039]

I searched for GEO series with human samples with the search terms "bone metastatases" and ended up with 244 results. I've included the series I thought are relevant. The entries in red initially looked relevant, but weren't because they were cell lines or non-human.

I also searched GEO series with human samples with the search terms "normal osteoblast" and found one other series that was not in the "bone metastatases" results that may be useful.

[ddolbae01]

I attached a revised excel file.
I reviewed the excel file that Lauren made and sorted the GEO data.

I defined indications according to groups (BM, BC, O) as followings.
* BM (bone metastasis from breast cancer)
1. sample from human female bone.
2. diagnosis as bone metastasis from breast cancer.
3. if the samples were obtained after chemotherapy, the samples were excluded.
* BC (Breast cancer without metastasis)
1. sample from human female breast.
2. Diagnosis as breast ductal carcinoma without any metastasis
or primary breast cancer cases among matched samples
3.The samples should not have metastasis at the time of biopsy of breast cancer
* O (normal human osteoblast)
1. cell line differentiated from human MSC (mesenchymal stem cell)
2. cell from bone tumor excluded.

Therefore, The numbers of BM, BC, and o GSMs were 79,44, and 21, respectively.

From Sungbae Park

[lsand039]

I'm not sure which samples from GSE51232 you want to use. The replicates seem to have different time points and progenitors.
There may be some samples from GSE86988 taken from the Ductal component of invasive breast carcinoma you may want to use. I have these listed under suggested Additions in the BC (PARK) tab.
I'm a bit confused as to whether you want to include the Osteoblast samples in GSE29036; these samples are not listed in the O (PARK) tab.
Should the normal bone marrow sample in GSE2361 be included? This sample is not listed in the O (PARK) tab.

Here are the series that will work easily with Efrain's code:
GSE57925
GSE51232
GSE46141
GSE39494
GSE38091
GSE34485
GSE2361

[ddolbae01]

There are 3 kinds of groups (BM: Bone metastasis from breast cancer, BC: Breast cancer alone, O: normal osteoblast) in this study
I finished clean and matched working
BM group has 6 GSE filles, 79 GSM files and 9,913 matched gene symbols
BC group has 4 GSE files, 45 GSM files and 1,1454 matched gene symbols
O group has 8 GSE files, 17 GSM files and 1,0359 matched gene symbols.
I attached the excel file

I am planning to start working in Banjo in the help of Lauren.

[ddolbae01]

I finally made a file for moving to Banjo.
There are three groups in this study as Bone metastasis group (BM), Osteoblast group(O) and Breast cancer alone group(BC).

The number of commonly expressed genes between BM and O is 8,690.
The number of commonly expressed genes among BM, O and BC is 7,006.
The number of specific genes that are expressed in BM and O, not expressed in BC is 1684.
I uploaded the file with list of genes.

Next step is to move to Benjo.

Thank you for Efrain and Lauren.

[ddolbae01]

Previous data file with common genes between BM and on, not BC have the gene information about only BM.
However, Lauren give the good idea about adding values of common gene in O group. And try to compare the result of BN between using GSMs data of BM and using GSMs data of O groups.
Because the BN using GSMs data of O group can provide physiologic interaction among the common genes, I think Lauren's advice more useful to get pathologic interaction among genes related with bone metastasis.
Therefore, I added the expression values of GSMs in osteoblast group into same file.

I attached the revised file.

[ddolbae01]

I summarized the process from late December, 2017 to today(Jan 22, 2018) as follows.

1. After discussion with Professor Yoo, I decreased the number or variable and increased the number of observation. (I deleted the non-important variables(genes) with missing data), Therefore, The number of variables and observation were 1219 and 56, respectively (1685-vars-49-observations in previous study). I attached the file as dataSmissing.txt. The genes of interests according to previous studies were TFGB3, PTK7 and ECT2 among the 1218 variables (except the group variable). The three genes are strongly related with bone metastasis of breast cancer. The three genes’ correlation scores were 0.1825, 0.3506, and, 0.2924, respectively.

2. I tried to the cytoscape program (version 3.6.0) to select the biologically meaningful genes and integrate the biological data and Bayesian network. The cytoscape can read the dot file made from Banjo and show the structure with arrow like banjo structure. There are many applications available in cytoscape that can reflect the biological data like as KEGG. However, the structure by using the app just reflected the biological data without no data of Bayesian network. Therefore. I tried to find the cancer related genes in the 1218 genes using a app as ClueGo. The ClueGo is a app. available to be installed in cytoscape and call the KEGG data network. I found 58 numbers of cancer related genes registered in KEGG using the ClueGo app. I have attached the file with 58 gene names and 3 genes of interest (TGFB3, PTK7, ECT2). The TGFB3 was included the 58 genes. Therefore, the number of genes was 60. The name of attached file is "integratedData60(KEGG).txt).

3.I analysed the 60 genes data and ran the Banjo three times each for 1 hr, 2hr, 4hr, 8hr, 16 hr and 24 hrs. The Banjo scores were as follows.
Banjo score /log normalization socre
path2 path3 path4
1hr -2861.5294 -2858.5609 -2861.1239
2hr -2859.441 -2854.7876 -2854.3111
4hr -2847.0231 -2854.2805 -2859.4661
8hr -2855.6711 -2851.905 -2847.9346
16hr -2842.512 -2848.3257 -2852.4438
24hr -2845.4011 -2847.6789 -2848.4833
36hr -2849.2855 -2841.4521 -2845.4067

1hr 1.36800027047785e-07 % 2.66249838356592e-06 % 2.05207200514178e-07 %
2hr 1.10424840152156e-06 % 0.00% 0.00%
4hr 0.2729557366 0.00% 1.07687671828871e-06 %
8hr 4.78975772556587e-05 % 0.00% 0.11%
16hr 24.84% 0.07% 0.00%
24hr 1.38% 0.14% 0.06%
36hr 0.03% 71.7045615810871 % 1.37%



In log normalization process, a banjo result with -2841.4529 had 71.7.

Now, If the banjo structure with -2841.4529 banjo score is best, I will get the 3rd degree of Markov blanket using 60 nodes. After that, I will try to reduced the number of node and simplify the structure using correlation score, influence score and Efrain order code.

[ddolbae01]

I posted the analytic result of a data file by using Efrain order coding.
The data has 9 variables.
I divided the 9 variables into three group (parent group, target group, and child croup) according to banjo results
The variables in parent group were 0, 1 and 2
The variables in target group was 3
The variables in child group were 4,5,6,7 and 8

The order as target-parent - child had the best score.

I attached the results file as Efrain code test.

[ddolbae01]

I got the cytoscape graphs (Hierarchic layout) based on banjo structure with best score.

I posted the 60nodes graphs, 1st neighbour nodes related with Group node, 2nd neighbour nodes related with Group, and 1st degree Markov blanket.

Red colour means Group node
Green coloured nodes are included nodes with top 10 correlation scores.
Purple coloured nodes mean the gene of interests.