Topological Data Analysis-Deep Learning Framework for Predicting Cancer Phenotypes

Lebohang Mashatola, Ismail Yunus Akhalwaya, Stephanie Muller

Keywords: Topological data analysis, Deep learning, Gene expression, Cancer Phenotype prediction

TL;DR: The use of topological data analysis to predict cancer-type phenotypes.

Abstract: Classification of patient cancer phenotypes from gene expression profiles remains a challenge in the field of transcriptomics. Gene expression data typically suffers from extreme noise and performs poorly on deep learning models alone. We build on recent work by Mandal et al., by incorporating the concept of differential gene expression analysis to pre-select genes that are necessary but not sufficient for disease association in our topological data analysis approach. The outcome is a reduction in computational cost in the calculation of persistent homology. We also test multiple topological representations to optimise prediction. Deep learning with topological features performs better compared to its use on raw data. Thus, topological features offers a new perspective on the difficult-to-unravel non-linear connection between genotype and phenotype

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Machine Learning for Sciences (eg biology, physics, health sciences, social sciences, climate/sustainability )