We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Biology > Molecular Networks

Title: Noncoding RNAs and deep learning neural network discriminate multi-cancer types

Abstract: Detecting cancers at early stages can dramatically reduce mortality rates. Therefore, practical cancer screening at the population level is needed. Here, we develop a comprehensive detection system to classify all common cancer types. By integrating artificial intelligence deep learning neural network and noncoding RNA biomarkers selected from massive data, our system can accurately detect cancer vs healthy object with 96.3% of AUC of ROC (Area Under Curve of a Receiver Operating Characteristic curve). Intriguinely, with no more than 6 biomarkers, our approach can easily discriminate any individual cancer type vs normal with 99% to 100% AUC. Furthermore, a comprehensive marker panel can simultaneously multi-classify all common cancers with a stable 78% of accuracy at heterological cancerous tissues and conditions. This provides a valuable framework for large scale cancer screening. The AI models and plots of results were available in this https URL
Comments: 7 pages and 3 figures
Subjects: Molecular Networks (q-bio.MN); Machine Learning (cs.LG)
Cite as: arXiv:2103.01179 [q-bio.MN]
  (or arXiv:2103.01179v2 [q-bio.MN] for this version)

Submission history

From: Anyou Wang [view email]
[v1] Mon, 1 Mar 2021 18:20:45 GMT (491kb)
[v2] Fri, 26 Mar 2021 18:06:45 GMT (1359kb)

Link back to: arXiv, form interface, contact.