No description, website, or topics provided.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Deep Induction

Maybe DeepDuction...

Project by causalnucleotidenetwork at the SVAI 2018 Hackathon P1RCC.

The Idea

Thought up in a whirlwind of ideas in 2-hours with insightful minds of:

When interacting with patients researchers/doctors/clinicians have access data in various forms:

  • Observations
  • NGS
  • EHR
  • etc...

However information isn't always complete or fully populated. This project seeks to tackle the problem of sparsity in this problem space. Specifically our problem statement is.

Given some information about the patient can we infer phenotypic or even genotypic data

Take a look at our first steps and hopefully we spark some insights and conversations as we work together towards a cure.

Method Overview

Given some information about the patient can we infer phenotypic or even genotypic data

We view this problem statement as a form of inductive reasoning. At a high level our approach follows suite:

  1. Learn the complex structure of Papillary Kidney Carcinoma patients
  2. Cluster patients to structure our complex data
  3. Glean scientific insight into these clusters A. Similar phenotypes or genotypes
  4. Acquire observations/data on a new patient
  5. Classify them in a cluster
  6. Inductively deduce additional features of the patient based on their assigned cluster

Learning Complex Structure

We employ a variational autoencoder to learn structure of p1RCC patients using gene expression data. The technique is originally form a paper by Gregory P. Way and Casey S. Greene.

Our notebook(s) contains the code and some additional documentation for our VAE implementation, again heavily inspire by Gregory P. Way and Casey S. Greene.


Primary source of data as TCGA

  • FPKM gene expression results (Clemson's PanTCGA Expression Data)
  • Clinical observations (xml) files

This repo contains module to parse, link, process clinical and FPKM data.


Extracting a Biologically Relevant Latent Space from Cancer Transcriptomes with Variational Autoencoders Gregory P. Way, Casey S. Greene bioRxiv 174474; doi:

Special thanks to the SVAI team for putting together such a great event!