Illuminating the “Black Box” – Biological Discovery with Deep Learning

October 19, 2018

Presented at SLAS HCS, Leiden, Netherlands

Deep Learning is poised to become a disruptive technology in drug discovery, e.g. for accelerated analysis in high-content screening [1, 2]. However, it often leaves project scientists with an uneasy feeling due to its black-box nature; they can’t retrace the network’s decisions and thus miss the insight they desire for signing off on the results.

In this poster we introduce a new approach – making the experimentalist “see” the findings of a deep learning network. We show methods to visualize and reason about the distinguishing aspects of cell images (morphologies, localized intensity differences …), e.g. separating control groups from each other. We illustrate exploration and validation of the classification process, shedding light into the “black box”. We also discuss how this approach may support new biological discoveries: Applied artificial intelligence can automatically identify new phenotypes in a truly unbiased way to catalyze R&D projects dealing with complex biology. 

[1] Kraus OZ, Grys BT, Ba J, Chong Y, Frey BJ, Boone C, & Andrews BJ (2017). Automated analysis of high‐content microscopy data with deep learning. Molecular systems biology, 13(4), 924.
[2] Godinez WJ, Hossain I, Lazic SE, Davies JW, & Zhang X (2017). A multi-scale convolutional neural network for phenotyping high-content cellular images. Bioinformatics, 33(13), 2010-2019.



Back to list