Interpretable deep learning uncovers cellular properties in label-free live cell images that are predictive of highly metastatic melanoma.

Autor: Zaritsky A; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA; Department of Software and Information Systems Engineering, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel. Electronic address: assafza@bgu.ac.il., Jamieson AR; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA., Welf ES; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA., Nevarez A; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA; Section of Molecular Biology, Division of Biological Sciences, University of California San Diego, 9500 Gilman Drive, San Diego, La Jolla, CA 92093, USA., Cillay J; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA., Eskiocak U; Children's Research Institute and Department of Pediatrics, University of Texas Southwestern Medical Center, Dallas, TX 75390, USA., Cantarel BL; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA., Danuser G; Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, Dallas, TX 75390, USA. Electronic address: gaudenz.danuser@utsouthwestern.edu.
Jazyk: angličtina
Zdroj: Cell systems [Cell Syst] 2021 Jul 21; Vol. 12 (7), pp. 733-747.e6. Date of Electronic Publication: 2021 Jun 01.
DOI: 10.1016/j.cels.2021.05.003
Abstrakt: Deep learning has emerged as the technique of choice for identifying hidden patterns in cell imaging data but is often criticized as "black box." Here, we employ a generative neural network in combination with supervised machine learning to classify patient-derived melanoma xenografts as "efficient" or "inefficient" metastatic, validate predictions regarding melanoma cell lines with unknown metastatic efficiency in mouse xenografts, and use the network to generate in silico cell images that amplify the critical predictive cell properties. These exaggerated images unveiled pseudopodial extensions and increased light scattering as hallmark properties of metastatic cells. We validated this interpretation using live cells spontaneously transitioning between states indicative of low and high metastatic efficiency. This study illustrates how the application of artificial intelligence can support the identification of cellular properties that are predictive of complex phenotypes and integrated cell functions but are too subtle to be identified in the raw imagery by a human expert. A record of this paper's transparent peer review process is included in the supplemental information. VIDEO ABSTRACT.
Competing Interests: Declaration of interests The authors declare no competing interests.
(Copyright © 2021 The Authors. Published by Elsevier Inc. All rights reserved.)
Databáze: MEDLINE