Brain, Lung and Bone metastasis of Breast cancer

[ddolbae01]

Now, I will post the process of study about metastasis of breast cancer.

1. Data enrollment

I searched the GSE data with clinical information as age, ER, PR, HER2 and adjuvant therapy.
Therefore, I enrolled 365 GSMs in 4 GSEs (GSE 2304, GSE 2603, GSE 5327 and 12276.
The enrolled GSMs have 13299 variables (10 clinical information as age, metastasis, bone metastasis, lung metastasis, brain metastasis, ER, PR, HER2, adjuvant chemotherapy and adjuvant hormone therapy and 13280 commonly expressed genes)

I attached two excel files that have raw data, table and data in enrolled GSMs.
I will move to 1st Banjo analysis at different hours (3hr, 6hrs, 12hrs, 24 hrs and 36 hrs)

[ddolbae01]

Now, I made the data file to be used in the 1st Banjo analysis.
I define the 1st data file including gene data of enrolled GSEs.
And, the 2nd data file including 10% genes in D0(1323), clinical information (10), and gene of interests (144) as D1 (1477).
(Gene of interests are composed of genes in signature genes, presented gene in GSE related studies and the three genes (NRF1, MYC, E2F4) recommended by professor Roy.)

The D1 data was composed of 365 subjects and 1477 variables.
So, I attached the D1 data as excel file and word file that was made by professor Roy.

Sincerely,

[ddolbae01]

I performed the 1st Banjo analysis with 1477 variables (1467 genes) and 365 observations for 1 hr, 3 hrs, 6 hrs, 12hrs, 24hr and 36hrs.
Each time analysis was implemented three times.
Among the 18 analyses, the network for 12hr analysis had the best score.
In addition, there was 12 genes within 3rd degree of Markov Blanket (MB).

Therefore, I made a data file (D2) for 2nd Banjo analysis.
D2 file was composed the 12 genes (3rd degree of MB), 156 signature genes, 3 genes (Roy recommended) and 10 clinopathological information.
Among 12 genes, 4 genes were overlapped among 156 signature genes.

Finally, the D2 data file had 177 variables that consisted of genes in 3rd degree MB, genes of interest and cliniopathological information.

I attached the results of 1st Banjo analysis as D1 data file, the Bayesian network with the best score, and 3rd degree of MB and D2 file.

[ddolbae01]

I performed 2nd Banjo analysis using 177 variables (10 clniopathological variables and 167 gene variables)
From 1hr analysis to 24hrs analysis (three times per each time), I got the best structure at 6hr analysis.
Therefore, I ordered the Banjo analysis for 36 hours, three times with initial structure of the 6hr analysis result.
Finally, I got the 2nd Bayesian structure (S2) with best score (-33520.784, 80.64%).

In 3rd degree Markov Blanket (MB), 49 variables were chosen: Metastasis, Bone metastasis, Brain metastasis, lung metastasis, Estrogen R, Progesteron R, HER2, Chemotherapy and 41 genes (signature genes)

In GeNIe structure, the accuracy of all 49 nodes was 0.836623 and accuracy of metastasis was 0.953425 (AUC = 0.88961 for metastasis =0 and 1)

In another analysis using prediction code, prediction for metastasis =1 was over 98% in almost cases.

Now, I am planning to use Efrain code (official name ?) and Ghong structure code. However, because the small number of variables is acceptable to use the Efrain code yet, I have to reduce the number of variables below 20 to use the Efrain code.
Therefore, I'd like to get the comment from professor Yoo and Roy.

I attached the results of 2nd Banjo analysis

[ddolbae01]

I posted the 2nd degree of MB previously.
In the previous structure, bone metastasis, brain metastasis and lung metastasis nodes were linked to metastasis nodes as " v" shape.
It means that the it is impossible the bone, brain, lung metastasis never do not happen simultaneously. It does not make sense.
Therefore, professor Yoo recommended me reanalysis without metastasis node.

I reanalyzed without metastasis node using D1 data (1476 variables = 1323 genes +9 clinical information + 144 genes of interests).
I got the structure with best score (S1, -368217.932, 75%) in 1st banjo analysis

In 2nd banjo analysis, I am using D2 data (139 variables = 3 metastasis nodes + 6 clinical information + 37 genes in 3rd degree MB + 93 genes of interests)
I will post the result up.

~~~~~~~~~~
Following professor yoo's recommendations, I am planning to subanalyses like as.
Analysis 2: No- Metastasis vs only Bone metastasis (exclude subjects who has Bone metastasis and brain and/or lung metastasis).
No-Metastasis: 228 subjects (among 365)
Only Bone metastasis: 77

Analysis 3: No- Metastasis vs only Brain metastasis (exclude subjects who has Brain metastasis and bone and/or lung metastasis).

No-Metastasis: 228 subjects (among 365)
Only Brain metastasis: 8

Analysis 4: No- Metastasis vs only Lung metastasis (exclude subjects who has Lung metastasis and bone and/or brain metastasis).

No-Metastasis: 228 subjects (among 365)
Only lung metastasis: 25

sincerely

[ddolbae01]

I will plan to do subgroup analyses - bone, brain and lung
And, I've performed the subgroup analysis - bone. (No- Metastasis vs only Bone metastasis (exclude subjects who has Bone metastasis and brain and/or lung metastasis)).

#1. Before analysis
Subjects: 305= 228 +77
No-Metastasis: 228 subjects (among 365) / Only Bone metastasis: 77
Variables
Total (1474, D1) = 10% genes (1323) + clinical information (7, exclude brain and lung metastasis) + gene signatures (144)

#2. 1st Banjo analysis
1. Among three times analysis per 3hrs, 6hrs and 12hrs, one of 12hrs results had the structure with best score
2.And, then I performed the three times analyses of 24hrs with initial structure as the 12hr result.
3. One of three 24 hrs, I got the structure (S1) with best score (-306718.1964 / 99.6122917430033%)

#3. 2nd Banjo analysis
Total (235, D2) = upper 10% ranked genes (correlated with bone metastasis in D1) (123)+ 3rd degree MB in S1 (14) + clinical information (7) + gene of interests (95) – overlapped genes (6) with genes within 3rd MB + gene of interest within 1st MB of bone metastasis in whole study (2)

#3. result of 2nd Banjo analysis - S2 (17 nodes)
- The structure is attached as "BCmeta3rd(bone)
- Genes in yellow colored node are gene of interests. Genes in green are clinical information.

#4. simple structures
- consist of nodes within 3rd MB of bone metastasis in S2
Let me show the structure as attached image as 3rdMB(bone)

#5. GeNIe Structure and prediction
- I attached GeNIe structure and CV file (GeNIeBone) for GeNIe analysis.
- the file with the results of prediction were attached (GeNIeBone.vsv.pre17)