CSE185 The Super Naïve Low Dimensional Embedding Clustering Method

This is group 16 final project for CSE185. It follows the algorithm description in the original t-SNE paper to implement t-SNE tool. The main application of TeeSnee is to cluster gene expression data of cells to reveal cell types.

Table of contents

• Sample Data Download
• Installation Instructions
• Basic Usage
• Complete Usage
• Benchmarking Information
• Contributors

Sample Data Download

BACK TO TABLE OF CONTENTS

Sample data (count matrix) taken from: https://www.10xgenomics.com/resources/datasets/human-brain-cancer-11-mm-capture-area-ffpe-2-standard

Check file data_processing.py to copy the commands needed to process the data to get gene by cell matrix.

Installation Instruction

BACK TO TABLE OF CONTENTS

• Install teesnee program with the following command:

git clone https://github.com/m1ma0314/CSE185Group16_TeeSnee.git

cd CSE185Group16_TeeSnee

• Python Installation Click Me to Get Python

• Pip Installation

python -m ensurepip --upgrade
python -m pip install --upgrade pip

• Installation requires the numpy, pandas, matplotlib, scikit-learn libraries to be installed. You can install these with pip:

pip install -r requirements.txt

• Change permissions of teesnee.py:

chmod 777 teesnee.py

Basic Usage

BACK TO TABLE OF CONTENTS

The basic usage of teesnee is:

python teesnee.py [-p targer_perlexity] [-z ifzipped] [-o output] filename

To run teesnee on a small test example (using files in this repo):

python teesnee.py -p 100 -o ./ minimal_dataset.csv

Complete usage instructions

BACK TO TABLE OF CONTENTS

The only required input to my_tsne is a cell x gene matrix data file. Users may additionally specify the options below:

• -p PERPLEXITY, --target_perplexity PERPLEXITY: specify target perplexity. If specified, the tsne function will calculate similarities matrix based on specified perplexity value and generate t-SNE plot. Higher perplexity value is associated with tighter clusters in the final output plot. Otherwise, the tsne function use perplexity=100 by default.
• -z ifzipped, --zipped: unzip dataset file if this argument is specified. By default, the datafile is viewed as unzipped and will be converted into matrix for further processing.
• -o FILE, --output FILE: Write output to file. By default, output is written to tsneplot.png

File Format

The output t-SNE plot is in png format.

Benchmarking Information

BACK TO TABLE OF CONTENTS

The benchmarking analysis is recorded in the folder benchmarking. Here is the time complexity plot comparison between teeSnee time complexity and scanpy’s t-SNE.

p.s. teeSnee still needs optimization for runtime TAT

Contributors

BACK TO TABLE OF CONTENTS

This repository was generated by Annabelle Coles and Mijia Ma, under the guidance of the original t-SNE paper and with inspiration from the article t-SNE from scratch.

Special thanks to CSE185 Professor Dr. Melissa Gymrek, TA Ryan Eveloff, TA Luisa Amaral, TA Himanshu.

Please submit a pull request with any corrections or suggestions. Your suggestions matter a lot to us <3