DeepBipolar: Identifying genomic mutations for bipolar disorder via deep learning

Background

DeepBipolar was the solution to the 2017 “Bipolar Exomes” challenge conducted by the Critical Assessment of Genome Interpretation 4. Proposed by the NSF Centre for Big Learning at the University of Florida, DeepBipolar was the most successful model submitted to the challenge, as per the assessments of the official assessors.

Problem Statement

Based on the dataset provided by the organisers, the Bipolar Exomes challenge involved identifying and distinguishing individuals who were affected by bipolar disorder from the unaffected. Within the realm of machine learning, this constituted a “supervised classification” learning problem.

My Contributions

Over the course of the efforts, I worked on:

Researching, designing, implementing, and testing the baseline Random Forests and Decision Trees models using traditional machine learning (ML) techniques.
Architecting the machine learning pipeline from data preprocessing, feature engineering, model training & testing, to model evaluation.
Conducting model hyperparameter experiments to understand the best hyperparameters that exhibited the best model performance
Testing & plotting different performance metrics like Precision-Recall and AUC-ROC curves to differentiate DeepBipolar from the traditional ML approach
Authoring the entire DeepBipolar paper based on in-depth collaboration with my teammates for understanding the proposed solution and all the technical details of the model
Iteratively improving the overall quality of the research paper by frequently collecting and incorporating feedback from my lab supervisor and teammates.

DeepBipolar performance compared to the other submissions

Impact

DeepBipolar was awarded as the most successful solution to the Bipolar Exomes challenge.
The DeepBipolar research paper was successfully published in the prestigious Human Mutation journal, therefore contributing to the medical community at large.
The DeepBipolar paper, as a research publication, is permanently publicly available (online) within the prestigious National Library of Medicine and Wiley publications.
The DeepBipolar paper currently enjoys 26 citations (Google Scholar) and 20 citations (Wiley Publications), cited by several medical research teams!

Project Links

Link to the Paper: DeepBipolar

How to cite this article:

PubMed

Sundaram L, Bhat RR, Viswanath V, Li X. DeepBipolar: Identifying genomic mutations for bipolar disorder via deep learning. Hum Mutat. 2017 Sep;38(9):1217-1224. doi: 10.1002/humu.23272. Epub 2017 Aug 1. Erratum in: Hum Mutat. 2018 Sep;39(9):1299. PMID: 28600868; PMCID: PMC5656045.

Wiley

Laksshman S, Bhat RR, Viswanath V, Li X. DeepBipolar: Identifying genomic mutations for bipolar disorder via deep learning. Human Mutation. 2017;38,1217-1224. https://doi.org/10.1002/humu.23272