I, for one am glad to see that when I'm not the only one who is at a loss to illustrate deep learning models and resorted to a commercial AI to make some squiggly lines (for the github tutorials).
You probably want to read this first before starting to install it!
Is this the thing that can take my metabolomics and proteomics of the same exact 800 or so tissues and integrate them properly? ...Not out of the box....
The current input categories are data that is already processed through R packages such as PharmacoGX and the single cell transcript measurements were processed with Seurat. This latter I find compelling enough to press the publish button, however. Thanks to work from Laurant Gatto's group (which should be out by now, I think, preprint link in here). I can get my single cell proteomics data into historical single cell formats. In my notes from a workshop he taught at EuBiC I wrote that he said "there is no reason we couldn't use this for other large -omics data" and also "omg, I'm way too dumb to be in this workshop" which I don't remember him saying, so was probably some level of intraspection on my part. (I miss when blogger had a spell checher embedded.... 😉)
Yes, the tool is not designed to do data processing for omics data. It expects data matrices in tabular format, where, in your case, you would have measurements from multiple metabolites and multiple proteins/peptides from matchings samples/patients with additional metadata about the patients (e.g. clinical variables such as survival, drug response, tumor stage etc). And, no, this tool is not designed for single-cell data processing either. It is designed for bulk omics datasets, however, if you provide processed single-cell data (as a matrix) it will also work, but we recommend other single-cell pipelines for dedicated single-cell analysis.
ReplyDeleteOf course, data processing is very important first step for omics analysis, but integrating processed data is another big challenge, which is what we aim to do with this package.
PS: The squiggly lines in the graphical abstract are not AI-generated. They are icons from the flaticon website, as far as I know they are hand-crafted. You can check https://www.flaticon.com/search?word=neural%20network
But, I agree it is hard to find a concise icon that represents different flavors of neural networks.
Thanks for your interest in the paper!