cognoma / cognoml

Machine learning package for Project Cognoma

Project Cognoma

Putting machine learning in the hands of cancer biologists.

Project Cognoma is an open source project to create a webapp for analyzing cancer data. We're a community-driven philanthropic project that began as a collaboration between the Greene Lab, DataPhilly, and Code for Philly. Our contributors are primarily based in the Philadelphia area, but anyone anywhere is welcome. This GitHub repository is the administrative and informational home of Cognoma.

The Meetup phase of Cognoma is now complete! The Childhood Cancer Data Lab of Alex's Lemonade Stand Foundation will be providing longterm maintenance. Public contributions are still welcome through GitHub. The main priority is enhancements and bug fixes to improve http://cognoma.org. For a nice overview of the project, see its coverage by The Philadelphia Citizen.

Teams and Repositories

The project is composed of four teams with their own corresponding repositories:

Team Name	Repositories	Description
Cancer Data	cancer-data, genes, figshare	processing the underlying cancer data to the formats required for this project.
Machine Learning	machine-learning, cognoml	building classifiers to predict mutation status from gene expression data.
Backend	core-service, task-service, ml-workers, infrastructure	creating the infrastructure to power the webapp and glue the components together.
Frontend	frontend, uiux	building the webapp that users interact with.

New Here?

If you are a new user and would like to get involved, please introduce yourself. Contributions are made through GitHub, so if you are unfamiliar with git or GitHub, check out the sandbox for a place to learn by doing.

Meetup Schedule

We hold project meetups. Our usual meeting spot is at Industrious (where CandiDate is located). The address is 230 S Broad St, Floor 17, Philadelphia.

📅 Date	⌚ Time	🗺 Location	ℹ️ Meetup Details	💰 Sponsor
Wednesday, October 11, 2017	6:00 PM	MilkBoy	DataPhilly	Alex’s Lemonade Stand Foundation
Tuesday, August 15, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, July 11, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, June 27, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, May 30, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, April 25, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, April 4, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, February 28, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Monday, February 13, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, January 31, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Monday, January 16, 2017	9:00 AM	Philly Think Space	Frontend Only	MLK Day Volunteers from Think Company
Tuesday, January 10, 2017	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, December 20, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, December 6, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, November 15, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, November 1, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, October 18, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, October 4, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Monday, September 19, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, September 6, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, August 23, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, August 9, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, July 26, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, July 19, 2016	6:00 PM	CandiDate	DataPhilly	Penn Institute for Biomedical Informatics
Tuesday, July 12, 2016	6:00 PM	CandiDate	DataPhilly	MilkBoy
Tuesday, July 5, 2016	6:00 PM	CandiDate	DataPhilly	Neo Technology
Tuesday, June 28, 2016	6:00 PM	MilkBoy	DataPhilly / Code for Philly	MilkBoy

Contributing

Community contributions are the driving force behind Cognoma. The heatmap below shows which users have contributed to which repositories:

See the guidelines for contributing for more information.

Maintainers

Cognoma relies on our generous community maintainers to assist with contributions. Thanks to the following maintainers for their help:

• Cancer Data: Claire McLeod (@clairemcleod)
• Machine Learning: Patrick Miller (@patrick-miller), Ryan Velazquez (@rdvelazquez), Jesse Prestwood-Taylor (@jessept), Yichuan Liu (@yl565)
• Backend: Derek Goss (@dcgoss), Andrew Madonna (@awm33), Kurt Wheeler (@kurtwheeler)
• Frontend: Benjamin Dolly (@bdolly)
• Community: Karin Wolok (@KarinSpiderwoman)
• Wildcards: Daniel Himmelstein (@dhimmel), Gregory Way (@gwaygenomics), Casey Greene (@cgreene)

cognoml's Issues

Write unit tests

Testing will include 3 different components

1. Testing for data
2. Testing for the classifier
3. Testing for the pipeline

Possible test frameworks to use: pytest, unittest, nose

Switch from printing to logging messages

The logging module will provide flexible management and storage for the messages which are currently printed out throughout the data reading and modeling process.

Add worker to query classifier data to fit cognoml model

@awm33 I created this so we can track it going forward.

A given classifier task has a list of entrezids and disease types. The worker code will query for any samples that match the list of disease types and join that to the mutations table. The result will not be an a [sample_id,mutation_status] form, so the worker needs to transform it into that form and pass it to the cognoml code.

Remove mutation filtering in CognomlData

Mutation data is taken directly from the front end, so this downloading/filtering can be removed.

Add cognoml Docker creation to this repository

Previously, we've played around with creating Docker images for running the cognoml package. See cognoma/machine-learning#62 (comment). We should probably create a docker directory in this repo and add Docker information and functionality there.

Improving figshare download modularity

Currently, there are three main issues we need to think about / address:

• cognoml/data.py and cognoml/figshare.py have duplicate functionality.

• We may want to make a subclass for FigshareDataSource in data.py. See #3 (comment)

• We need to find a way to expose figshare download functionality to other cognoma repositories. For example, this could help with cognoma/core-service#42

Use OrderedDict in performance processing

Dicts do not preserve ordering and the output is in JSON form, so we may end up seeing erroneous diffs if ordering is not preserved. This is just needed in the get_results method in the CognomlClassifier class.

Create base data class

CognomlData assumes too much about the size and shape of the data sets. We need to create a base data class with some basic methods, then let additional classes inherit those methods.

Explore optimization of machine learning pipeline

The current classifier pipeline takes a long time to fit and may be fitting the same model multiple times. This should be looked at in hopes of finding some low-hanging fruit in performance gains.

Write integration tests

We have a separate issue for unit tests, integration tests will be more complicated.

We need to create a "test" data set, run it, and check it against expected output.