Predicting Cancer Stage Using miRNA

Outline

Background

KIRC

Kidney Renal Clear Cell Carcinoma (KIRC) is the most common form of kidney cancer.

Data

TCGA

The Cancer Genome Atlas (TCGA) data portal was used to download data.

Methods

Algorithms

  1. Support vector machine (SVM)
  2. Random forest (RF)
  3. Naive Bayes (NB)

Preprocessing

  1. Filter based on intensity of expression and coefficient of variation (subsequently labeled PRE1)
  2. Filter based on p-value of t-test and transform using principal compenents (subsequently labeled PRE2)

Splitting

  1. 5-fold cross validation (subsequently labeled 5CV)
  2. 10-fold cross validation (subsequently labeled 10CV)

Results

Accuracy

plot of chunk unnamed-chunk-2

Specificity/Sensitivity

plot of chunk unnamed-chunk-3

Area Under Curve

plot of chunk unnamed-chunk-4

Analysis