Explainable Deep Learning for Augmentation of sRNA Expression Profiles

TL;DR

This study demonstrates that deep learning models significantly improve the prediction of metadata such as tissue, age, and sex from small RNA expression profiles, aiding in data annotation and biological interpretation.

Contribution

It systematically benchmarks deep learning against random forests for sRNA metadata augmentation, highlighting DL's superior performance and interpretability via feature importance analysis.

Findings

Deep learning achieves over 96% accuracy in tissue prediction.

DL outperforms RF in all tested metadata predictions.

Feature importance analysis reveals biologically relevant sRNAs.

Abstract

The lack of well-structured metadata annotations complicates there-usability and interpretation of the growing amount of publicly available RNA expression data. The machine learning-based prediction of metadata(data augmentation) can considerably improve the quality of expression data annotation. In this study,we systematically benchmark deep learning (DL) and random forest (RF)-based metadata augmentation of tissue, age, and sex using small RNA (sRNA) expression profiles. We use 4243 annotated sRNA-Seq samples from the small RNA expression atlas (SEA) database to train and test the augmentation performance. In general, the DL machine learner outperforms the RF method in almost all tested cases. The average cross-validated prediction accuracy of the DL algorithm for tissues is 96.5%, for sex is 77%, and for age is 77.2%. The average tissue prediction accuracy for a completely new dataset is 83.1% (DL) and 80.8% (RF). To understand which sRNAs influence DL predictions, we employ backpropagation-based feature importance scores using the DeepLIFT method, which enable us to obtain information on biological relevance of sRNAs.