alyssafrazee

import pandas as pd
from numpy.random import randint
from numpy import median, percentile

my_data = pd.read_csv('dataset.csv')
n = len(my_data)

num_bootstrap_samples = 1000
bootstrap_results = []
for b in xrange(num_bootstrap_samples):

alyssafrazee

Using Ballgown and Polyester

Ballgown

Reading data

• ballgown creates a ballgown object from tablemaker output
• ballgownrsem creates a ballgown object from RSEM output. (not yet well-tested).
• gffRead and gffReadGR read GTF (annotation) files into R
  • gffRead gives you a data frame
• gffReadGR gives you a GRanges object

alyssafrazee

## simple analysis code for GEUVADIS data
## AF Oct 2014

library(ballgown) #biocLite
library(RSkittleBrewer) #install_github
library(RColorBrewer) #CRAN
library(usefulstuff) #install_github
library(RCurl) #CRAN

load('fpkm.rda') # download at http://files.figshare.com/1625419/fpkm.rda

alyssafrazee

ideas for side projects

FOR LEARNING!

https://twitter.com/pgbovine/status/518171410402127872

alyssafrazee

starting a new project

1. Make a repository on GitHub. Check the box that says "initialize this repo with a README."
2. Clone that repository on to your computer. That's git clone + the ssh URL you can find on the right-hand side of the repository. Go to the directory where you want the repository_name folder to live.
3. Run a git status as a sanity check. (Everything should be clean).
4. Write some code!
5. Run a git status again. See that your code now lives in your repository, but hasn't yet been added to version control. (the file names should be red in your terminal)
6. "Add" (git add) the code to version control.
7. Run another git status. The file names should now be green, meaning they've been added to the version control staging area, but not committed to your repository's history.
8. Commit your changes with git commit -m 'message_here').

alyssafrazee

## create ballgown objects with GEUVADIS data

source("http://bioconductor.org/biocLite.R")
biocLite('ballgown')
library(ballgown)
system('mkdir -p Ballgown/small_objects')

## make phenotype table:
dataDir = 'Ballgown/' #tablemaker output lives here
sampnames = list.files(dataDir, pattern = 'H|N')

alyssafrazee

notes from BioC 2014

packages and the git-svn bridge

instructions on website I didn't get my invite to get an SVN account - mystery.

• add bioc-sync as a collaborator on your git repo
• then, webhooks & services --> add webhook. (there's a URL for this on the bioc instruction page.)
• git-svn bridge doesn't really do merging very well: it's "winner-take all." you have to pick whether git or svn wins on merge conflicts.
• only deals with master branch of git repo

alyssafrazee

San Francisco

breakfast

• Plow
• Mission Beach Cafe
• Griddle Fresh
• Toast

coffee

• Reveille
• Philz

alyssafrazee

# power calculation examples

get_power = function(truep, p0, n, alpha=0.05) {
    num_rejections = 0
    for(i in 1:10000){
        dat = rbinom(n, size=1, prob=truep)
        pv = 2*(1-pbinom(sum(dat), size=n, prob=p0))
        if(pv < alpha) num_rejections = num_rejections + 1
    }
    return(num_rejections / 10000)

alyssafrazee

interesting datasets on the internet

• Hilary Mason's collection of research-quality datasets
• 100+ Interesting Data Sets: seems really great for ML/data science practice or fun side projects.
• Most of the datasets available with R, but here ALSO available in CSV format! 700+ datasets.
• Kaggle Higgs Boson data (I think this is a super cool problem)
• Statistical Sleuth data problems -- pretty good for intro stats concepts
• Hadley's data packages -- baby names by sex (1880-2013), fuel economy of cars, atmospheric measurements from Central America, and info on all NYC flights in 2013. Set up for R, but I bet you could process these with other software if you wanted.
• others? (leave a comment!)