TCGA Biomarkers Identification using Machine Learning | Complete Walkthrough

Well, mostly doing this since people have been asking to connect the database with some basic machine learning script , so I might as well capitalized on this. Anyhow, I mostly wrote this with the mindset on education and not really on research so the code was keep as simple as possible. Despite the high accuracy, i don't think the markers identified from this script is going to be siper useful but i think if someone can try to run a deseq2 or limma on the dataset and compaerd the results, I would love to heard from that

Slides used

Script (look for TCGA_biomarkers)

Chapters:
0:00 Introduction and background
5:00 Chapter 1 - Installing packages and importing libraries
7:20 Using TCGA Biolinks
12:50 Structuring Input data and filtering
17:36 PlotMDS from limma and edgeR
18:19 Normalization of data
21:22 PCA Analysis
22:05 Making Train Label and One -hot Encoding
25:04 Chapter 2 - Neural network construction
30:53 Neural networking Training model fitting
33:31 Saving Model as hdf5 files
34:41 Extraction weights and bias
37:02 Extraction of GOI using weights and bias
41:06 Chapter 3 - Gene set enrichment analysis
44:53 Results!!!!!!
47:15 Some major issues with this approach

[Muuip]

Great presentation!

Much appreciated!👍

[rk]

Thanks again for previous help! One question I'm having issue with. I used legacy data, so the GDCprepare function did not work for my query. I was able to design a workaround where I created a count matrix manually, with the metadata in a separate table. How would I go about assigning training labels to data that's not in SummarizedExperiment format? I.e. how would I replicated the below code by using two separate data frames - one for expression data and another for metadata:

train_label <- train_label %>% as.factor() %>% as.numeric ()
train_label <- train_label - 1
dim(train_label) <- c(dim(expr_filtered)[2], 1)

[mangalahegde]

Hi, Thank you so much. This is really helpful

[md.ishtiakrashid]

I was wondering if this could be used in case of 16S amplicon sequencing data. Could you please enlighten me?

[Muuip]

Time series on binomial outcomes!?

Ex.: gene expression comparison by death versus survival over 30 days!?
🤔

[vngplex]

Hi I'm having a hard time understanding about finding goi. Can you tell me about the theory of finding good nodes (gene of interst) by summing total weight and bias?

[anthonyimhenkuomon]

Hi,
This is absolutely a great video!!
Please, I did not understand your explanation in code line 70: about removing "Metastatic". I am doing similar ML project with "TCGA-BLCA" i.e Bladder cancer but not with ANN. my sample group is "Primary Tumor, Solid Tissue Normal". I was thinking of how to implement code line 70, but, I actually don't get it.
could you brief me in line with my Project?

regards,

[RanjeetSingh-xwlr]

Hi there! Awesome video - super cool stuff! I ran the code

dge <- DGEList(t(expr_filtered))

and an error message popped up saying : 'Error in plot.xy(xy, type, ...) : invalid color name 'Primary Tumor'' - any clue as to what might have gone wrong and how I can fix? Thank you so much!

[MadihaHameedAwan]

you did not tel me how to select project id because manifest file downloading is tricky you firs need to tel that

[rutchristine]

hi I tried your code especially for PCA visualization but my graph would not show. It says table of extent 0>. How do I solve this? I am totally new in r. Thank you