GENIUS: GEnome traNsformatIon and spatial representation of mUltiomicS data
01:36 Share

Link to bioRxiv paper: http://biorxiv.org/cgi/content/short/2023.02.09.525144v1?rss=1

Authors: Sokac, M., Dyrskjot Andersen, L., Haibe-Kains, B., Aerts, H. J. W. L., Juul Birkbak, N.

Abstract: Deep learning is widely used in many applications including medical imaging, speech recognition, and language processing. Those models have often been considered black boxes and are not popular in multi-omics research because they lack interpretability. The multi-omics data often results in large quantities of observation in tabular form, where we do not assume spatial connectivity between observations. In most cases, using this type of data is computationally heavy on statistical analyses, which have to be corrected for a false discovery rate, leading to a lot of discarded results. Here, we present a framework with multiple options for integrating multi-omics data by transforming the data into new spatially dependent space which can be used for inference. The framework we developed is able to transform multi-omics data into images and successfully extract relevant information with respect to the desired output. Our results are focused on the prediction of metastatic cancer without losing the ability of model inference, resulting in the top 10 genes which are associated with metastatic disease progression in six cancer types included in this study. We anticipate our framework to be a starting point for a multi-omics data transformation and analysis without the need for statistical correction.

Copy rights belong to original authors. Visit the link for more info

Podcast created by Paper Player, LLC