Evaluate performance of covariates at predicting various mutations #47
Conversation
Creates an explore directory and README for this type of exploratory notebook. See how well covariates (non-expression features) predict TP53 mutation. Related to cognoma#8: General mutation-load does provide some ability to predict mutation status of TP53. Partially addresses cognoma#21: Covariates are extracted from samples.tsv.
This isn't too surprising given TP53's role in controlling cell cycle checkpoints. Is this only true for genes in cell cycle checkpoint or DNA repair pathways, or is it also true for genes in proliferation pathways? |
Evaluate covariate-only classifiers for the interesting mutations compiled in cognoma/cancer-data#22 (comment). Switches to an expand grid system for evaluating all possible covariate combinations. Plot performance of all covariates on each mutation. Switches to `covariates.tsv` created in cognoma/cancer-data#24 for encoded covariates.
@cgreene, I updated the notebook to evaluate performance of covariate-only classifiers for the 8 interesting mutations we've previously considered. Here is performance of the models with all covariates included:
So actually, TP53 is among the hardest to predict using only covariates. VHL which is highly disease-specific achieves a near-perfect AUROC. Therefore, without the disease or organ covariate, expression classifiers of VHL are likely just classifying kidney clear cell carcinoma / kidney tissue. See this dataframe to get a general idea of covariate importance. Mutation load and disease type both seem important. |
|
@dhimmel : Interesting! I can imagine that gene expression would clearly capture disease/organ, so if one usually had a mutation in a single gene that was relatively specific (e.g. VHL) I could imagine that creating a strong signal. |
dhimmel
changed the title
| @@ -0,0 +1,280 @@ | |||
|
|
|||
There was a problem hiding this comment.
looks like you need to nbconvert this again
There was a problem hiding this comment.
I'll take a look again once it's easier to read
There was a problem hiding this comment.
Good catch. should be fixed now.
| @@ -0,0 +1,9 @@ | |||
| # A directory for exploratory machine learning analyses | |||
|
|
|||
| This directory is home is exploratory analyses that help answer questions about how we should do machine learning. For algorithm implementations see the [`algorithms`](../algorithms) directory. For other types of analyses, place them here. | |||
There was a problem hiding this comment.
"This directory is home to exploratory analyses"?
|
|
||
| # coding: utf-8 | ||
|
|
||
| # # Create a logistic regression model to several mutations from covariates |
There was a problem hiding this comment.
"Create a logistic regression model to predict several mutations from covariates"?
| return pd.DataFrame.from_records(rows, columns=data_dict.keys()) | ||
|
|
||
| mutations = { | ||
| '7157': 'TP53', # tumor protein p53 |
There was a problem hiding this comment.
pep8 says inline comments should be separated by two spaces
| # In[8]: | ||
|
|
||
| def get_aurocs(X, y, series): | ||
| X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=0) |
Creates an explore directory and README for this type of exploratory notebook.
See how well covariates (non-expression features) predict TP53 mutation.
Related to #8: General mutation-load does provide some ability to predict mutation status of TP53.
Partially addresses #21: Covariates are extracted from
samples.tsv.