Learning statistics: Principal Component Analysis

We decided to apply Principal Component Analysis for upcoming multisample BAM QC analysis in Qualimap.

The idea is very simple: we compare 2 and more BAM files based on a number of various metrics including mean coverage, GC content, insert size, mapping quality and others. Since there are multiple comparison features avaialable one can apply PCA to analyze how different are the samples from each other by using biplot.

Here is a small collection of tutorials on PCA that I went through to get more confident with the topic.

1) An intuitive explanation of PCA
Nice overview of the topic, focusing on understanding the application of PCA.

2) A tutorial on principal component analysis
Practical tutorial with very low entry level which allows to do try PCA on a small example

3) Using R for Multivariate Analysis
Great tutorial on multivariate analysis using R. Totally recommended.

Learning statistics: Simpsons paradox

Although I had a couple of nice courses in the university, I still feel myself not very confident when going through the statistics analysis part in a genomics paper. Apparently, it is not enough to know algorithms and biology to become a bionformatician :) After realizing (finally!) that statistics is super important for science and especially for data analysis, I started taking some MOOCs (this,this and this for example ) to refresh and improve my knowledge. As a result I am learning a lot of cool new things now! :)

Today I came across something interesting in the Exploratory Data Analysis course: Simpson's paradox.

Somehow I never heard about it before or just did not keep attention to it. The idea is the following: main trend or correlation is different for distinct groups in the dataset compared to the dataset as a whole. Such effect is usually introduced by some unrecognized confounding factor in the data.

Nice visualization and explanation can be found here.