RT Journal Article
SR Electronic
T1 Deep convolutional neural networks allow analysis of cell motility during stem cell differentiation and neoplastic transformation
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 2017
UL http://biorxiv.org/content/early/2017/07/12/159202.abstract
AB Cells in culture display diverse motility behaviors. In multiple contexts, motility behaviors reflect broader cell 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 a class of 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, CNNs are a promising approach to learn relevant features for motility analysis from data, rather than requiring a domain expert to engineer features by hand. Here, we apply CNNs to classify different motility behaviors by representing motility as a 3D space with markers denoting a cell’s location at each time point. 3D CNNs provide accurate classification of several simulated motility behaviors, the motility behaviors of multiple cell types, and characteristic motility behaviors of commitment states in myogenic cells. Autoencoders were trained effectively to learn representations of these 3D motility spaces in an unsupervised manner. We show that this approach can achieve reliable detection of differentiation state for muscle stem cells and better-than-chance detection of neoplastic transformation in a cancer cell model. The variety of cell type differences we can detect suggests that the algorithm is generally applicable to novel cell types. While we have applied these methods to the analysis of cell motility, our scheme for representing motion spatially for analysis by CNNs is generalizable to other motion classification problems.