A Systematic Approach to Featurization for Cancer Drug Sensitivity Predictions with Deep Learning
Authors:
Austin Clyde,
Tom Brettin,
Alexander Partin,
Maulik Shaulik,
Hyunseung Yoo,
Yvonne Evrard,
Yitan Zhu,
Fangfang Xia,
Rick Stevens
Abstract:
By combining various cancer cell line (CCL) drug screening panels, the size of the data has grown significantly to begin understanding how advances in deep learning can advance drug response predictions. In this paper we train >35,000 neural network models, sweeping over common featurization techniques. We found the RNA-seq to be highly redundant and informative even with subsets larger than 128 f…
▽ More
By combining various cancer cell line (CCL) drug screening panels, the size of the data has grown significantly to begin understanding how advances in deep learning can advance drug response predictions. In this paper we train >35,000 neural network models, sweeping over common featurization techniques. We found the RNA-seq to be highly redundant and informative even with subsets larger than 128 features. We found the inclusion of single nucleotide polymorphisms (SNPs) coded as count matrices improved model performance significantly, and no substantial difference in model performance with respect to molecular featurization between the common open source MOrdred descriptors and Dragon7 descriptors. Alongside this analysis, we outline data integration between CCL screening datasets and present evidence that new metrics and imbalanced data techniques, as well as advances in data standardization, need to be developed.
△ Less
Submitted 4 May, 2020; v1 submitted 30 April, 2020;
originally announced May 2020.
