RT Journal Article
SR Electronic
T1 Deep convolutional and recurrent neural networks for cell motility discrimination and prediction
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 159202
DO 10.1101/159202
A1 Jacob C. Kimmel
A1 Andrew S. Brack
A1 Wallace F. Marshall
YR 2018
UL http://biorxiv.org/content/early/2018/02/09/159202.abstract
AB Cells in culture display diverse motility behaviors that may reflect differences in cell state and function, providing motivation to discriminate between different motility behaviors. Current methods to do so rely upon manual feature engineering. However, the types of features necessary to distinguish between motility behaviors can vary greatly depending on the biological context, and it is not always clear which features may be most predictive in each setting for distinguishing particular cell types or disease states. Convolutional neural networks (CNNs) are machine learning models ideally suited to the analysis of spatial data, allowing for relevant spatial features to be learned as parameters of a model. Given that motility data is inherently spatial, we apply CNNs to classify different motility behaviors using two novel approaches. The first approach represents motility explicitly as a 3D space with markers denoting a cell’s location at each time point, and the second utilizes recurrent long-term short-term memory (LSTM) units to represent the temporal dimension implicitly. Both 3D CNNs and convolutional-recurrent neural networks (RNNs) provide accurate classification of simulated motility behaviors, the experimentally measured motility behaviors of multiple cell types, and characteristic motility behaviors of muscle stem cell differentiation states. The variety of cell motility differences we can detect suggests that the algorithm is generally applicable to novel cell and sample types. 3D CNN and RNN based autoencoders were also effectively trained using the explicit 3D representations to learn motility features in an unsupervised manner. Additionally, adapted RNN models effectively predict muscle stem cell motility from past tracking data.