Deep Neural Networks Identify Signaling Mechanisms of ErbB-Family Drug Resistance From a Continuous Cell Morphology State Space

It is well known that the development of drug resistance in cancer cells can lead to a change in cell morphology. We reasoned that machine-learning techniques could thus be used to elucidate far greater insight into the relationship between cell shape and signaling. To test this hypothesis we performed a large high content screen on drug sensitive and drug resistance cancer cells, and analysed the shape of these cells using a deep neural network. Our model identified a continuous 27-dimension space describing all of the observed cell morphologies from which we were able to predict drug resistance with an accuracy of 74%. In addition, analyzing changes in cell morphology identified signaling networks that, when perturbed, caused the death of drug resistant cells. These findings suggests that complex morphologies can decode states of signaling networks seemingly unrelated to cell shape, and that analysis of this information can unravel cellular mechanisms hidden to conventional measurements.