Rule-based cell annotation in Single-Cell RNA-Seq. Are we on right track?

The post is in Single Cell Biology (not changed the content much from LinkedIn):

Details & discussion can be found here: https://www.linkedin.com/feed/update/urn:li:activity:6540616360551161856

A fantastic pre-print that seemed to have touched a very important point of bias in rule-based cell annotation/assignment via selected marker genes. Something I always felt when I first saw single-cell RNA-Seq analysis in 2017. In 2019, it is amazingly transformative & helpful technology but still needs improvement as to my knowledge this problem persists. I will pen down my thoughts based on a LinkedIn post I made on this after reading an article

1. A work that pushed my critical thinking about the problem:

Fabian Theis group work led by Niklas D. Köhler & Maren Büttner points it out as well.

Deep learning does not outperform classical machine learning for cell-type annotation

Some of my thoughts linked to the pre-print, discussions with others & its future prospects are outlined below:

Link1: https://lnkd.in/gBvWbzM

Having said that, I also encourage another Twitter thread that is linked to this:

Link2: https://lnkd.in/gNk6H9r

Greg Finak makes an amazing point about how single cell community often misinterprets or misrepresents while conveying the high-dimensional plots & if we do not take care it will haunt us in the future. Encourage everyone to read the pre-print, thread & reflect to understand the points addressed both in cell assignment via annotation & how deep learning/machine learning aids in Single-Cell RNA-Seq space.

I would encourage everyone to read the series of discussion I had with Mike Huss in the LinkedIn thread. We touched upon some insightful exchange of ideas there. I cannot thank enough Mike Huss for the engagement. We both agree about the transfer learning approach based on what Human Cell Atlas(HCA) is trying to build across tissues and diseases so as to improve or deviate from rule-based cell annotation.

A new version of Seurat V3 is published on such lines using anchoring approaches but I will make a separate post on that publication after detailed reading of that work along with LIGER.

(Souce: https://satijalab.org/seurat/)

Two articles in that space that are worth reading for single cell multi-omics integration:

Article 1: https://www.cell.com/cell/pdf/S0092-8674(19)30559-8.pdf

Article 2: https://www.cell.com/cell/pdf/S0092-8674(19)30504-5.pdf (LIGER Github: https://github.com/MacoskoLab/liger)

#AI #DataScience #predictiveanalytics #healthcare #medicines #genomics #pharmaindustry