Chapter 8 Classification

8.1 Introduction

Imagine you have RNA-seq of a collection of labeled normal lung and lung cancer tissues. Given a new sample of RNA-seq from the lung with unknown diagnosis, will you be able to predict based on the existing labeled samples and the expression data whether the new sample is normal or tumor? This is a sample classification problem, and it could be solved using unsupervised and supervised learning approaches.

Unsupervised learning is basically clustering or dimension reduction. You can use hierarchical clustering, MDS, or PCA. After clustering and projection the data to lower dimensions, you examine the labels of the known samples (hopefully they cluster into separate groups by the label). Then you can assign label to the unknown sample based on its distance to the known samples.

Supervised learning considers the labels with known samples and tries to identify features that can separate the samples by the label. Cross validation is conducted to evaluate the performance of different approaches and avoid over fitting.

StatQuest has done an amazing job with machine learning with a full playlist of well organized videos. While the full playlist is worth a full course, for the purpose of the course, we will just highlight a number of widely used approaches. They include logistic regression (this is considered statistical machine learning), K nearest neighbors, random forest, and support vector machine (these are considered computer science machine learning).

8.2 Supervised learning

8.3 Cross validation

8.4 Regression

8.5 Regularization

8.5.1 Ridge regression

8.5.2 LASSO regression LASSO tutorial in R

8.6 KNN

8.7 Decision trees

8.8 Random forest

8.9 SVM

8.10 Lab 4

8.10.1 K-Nearest Neighbors tutorial

8.10.2 Regression/Ridge/LASSO Tutorial

8.10.3 Logistic Regression Tutorial

8.10.4 Support Vector Machine Tutorial

8.10.5 Random Forest Tutorial