danielzucha/ngs19

materials for #NGSchool2019 - Machine Learning for Biomedicine

Materials for #NGSchool2019 - Machine Learning for Biomedicine

You will find here the materials for workshops, hackathons and lectures at the #NGSchool2019, together with installation directions and tips for running the software necessary for participation in the NGSchool2019.

Table of Content

• General instructions
  • Colab
• Workshops
  • Intro to HPC
  • Intro to R
  • Intro to Python
  • Intro to Stats
  • Unsupervised learning
  • Bayesian Inference
  • Natural language processing
  • Reinforcement Learning
  • Deep learning methods for genomics
  • Deep Generative Models for dimensionality reduction
  • Tree based methods
  • Lasso workshop
• Hackathons
  • Dilated Convolutional Neural Nets for DNase-seq and ATAC-seq footprinting

General instructions

Colab

Google Colab is an online service in which you can run jupyter notebooks (and even use some limited GPU!) It comes with some preloaded libraries which makes it easier to teach and run tutorials without having to spend too much time on fixing dependencies etc.

Workshops

Intro to HPC

tutor: Klemens Noga

The wbsite with info about the workshop can be accessed here

Intro to R

tutor: Maja Kuzman

Intro to Python

tutor: Kasia Kędzierska

The whole workshop will be executed in the Jupyter notebook, and will rely on several Python packages. In the directory you can find a setup_check.sh script you can run to see if your enviorenment satisfies all requirements.

Install and check if requirements are satisfied.

bash intro_to_python/setup_check.sh

Requirements:

• python3
• Jupyter
• python3 modules:
  • numpy
  • pandas
  • matplotlib
  • scipy

Intro to Stats

tutor: German Demidov

Unsupervised learning

tutor: Kasia Kędzierska

Slids: unsupervised_learning/unsupervised_learning_slides.pdf

The workshop will be run in R notebook. We would work locally and the following packages are required.

Requirements:

• R 3.5+
• tidyverse 1.2.1+
• factoextra 1.0.5+
• ggpubr 0.2+
• ggsci 2.9+
• MASS 7.3-50+
• tsne 0.1-3+
• umap 0.2.3.1+

required_packages <- c("tidyverse", "factoextra", "ggpubr", 
                       "ggsci", "MASS", "tsne", "umap")

for (pkg in required_packages) {
  if(!require(pkg, character.only = TRUE, 
              quietly = TRUE, 
              warn.conflicts = FALSE)) {
    print(paste0("Warning! Installing package: ", pkg, "."))
    install.packages(pkg)
  } 
}

print("All done! :)")

Bayesian Inference

tutor: Roman Cheplyaka

Either in RStudio or in the interactive R session run following commands:

required_packages <- c("rstan", "StanHeaders", "magrittr", "reshape2", 
                       "forcats", "stringr", "dplyr", "purrr", "readr",
                       "tidyr", "tibble")

for (pkg in required_packages) {
  if(!require(pkg, character.only = TRUE, 
              quietly = TRUE, 
              warn.conflicts = FALSE)) {
    print(paste0("Warning! Installing package: ", pkg, "."))
    install.packages(pkg)
  } 
}

print("All done! :)")

Natural language processing

tutor: Noura Al Moubayed

Installation guidelines

1. Install miniconda

Start by installing miniconda.

https://docs.conda.io/en/latest/miniconda.html

2. Create conda environment

To simplify, we can crete the enviromnet from the yml file: nlp/workshop.yml

conda env create -f nlp/workshop.yml

3. FROM LOCAL COPY Install missing package:

Due to a large file size (>1GB), we are copying the en_core_web_lg from USB sticks distributed on site. When you copy the file from a USB, please change the following command to point to the location of the file.

# python -m spacy download en_core_web_lg
conda activate workshop
pip install /path/to/folder/with/en_core_web_lg-2.2.0.tar.gz

4. Clone the repository

Make sure your github repository is up to date and unpack one of files from the nlp directory! The files is gziped to reduce its size.

git pull origin master
gunzip nlp/tutorial_features.pkl.gz

Running the workshop

cd nlp
conda activate workshop
jupyter notebook

Reinforcement Learning

tutor: Robert Loftin

Deep learning methods for genomics

tutor: Ron Schwessinger

The seminar hands-on workshop will be run in a google colab notebook. A google account is required though. Additional information can be found in this repo but no need to install anything for the workshop.

Deep Generative Models for dimensionality reduction

tutor: Kaspar Märtens

Tree based methods

tutor: Rosa Karlic

Lasso workshop

tutor: Tim Padvitski

Hackathons

Dilated Convolutional Neural Nets for DNase-seq and ATAC-seq footprinting

Requirements:

• python3
• keras and Tensorflow v1.14 as backend
• numpy
• scikit-learn
• google account for colab notebook work

Literature:

• Dnase Footprinting general Introduction
• Learning genomic signals at base pair resolution