publication date: Oct. 30, 2020
Conversation with The Cancer Letter
Elodie Pronier, PhD
Translational research scientist, Owkin
Will an oncologist of the not-so-distant future be able to pull up an image of a tumor biopsy slide on a screen and—without having to order a biomarker test—see the molecular characteristics of the cancer?
Owkin, a company headquartered in New York City and Paris, is providing proof of concept that artificial intelligence is already capable of automating that process.
In a paper recently published in Nature, Owkin’s researchers make the case that their AI model, HE2RNA, can be trained to “systematically predict RNA-Seq profiles from whole-slide images alone, without expert annotation.” Owkin specializes in using machine learning to generate predictive models.
Using data from The Cancer Genome Atlas as well as from Owkin’s partner hospitals, the researchers tested the HE2RNA model in multiple cancer types and found statistically significant correlations across a broad swath of genes that are commonly expressed in cancers.
“An average of 3,627 genes (respectively 12,853), including 2,797 protein-coding (respectively 8,450) per cancer type were predicted with statistically significant correlation under [Holm–Šidák] correction (respectively under [Benjamini–Hochberg] adjustment,” the authors report in the Nature paper, published Aug. 3.
The researchers demonstrated that HE2RNA was able to:
Predict the expression of genes involved in cancer development in many cancer types. According to the paper, specific sets of genes—e.g. CD3 … Continue reading Owkin’s AI model provides proof of concept that machine learning can predict gene expression across cancer types
Institutional subscribers, please log in with your IP.
If you’re not a subscriber why not join today?
You will be given immediate access to premium content on the site.
Click here to join.
Credit: Google News