Quantifying the nuclear localisation of fluorescently tagged proteins

Cells are highly dynamic, continually responding to intra- and extracellular signals. Monitoring and measuring the response to these signals at the single-cell level is challenging. Signal transduction is fast, but reporters for downstream gene expression are typically slow, requiring fluorescent proteins to be expressed and to mature. An alternative is to fluorescently tag and then monitor the intracellular locations of transcription factors and other effectors. These proteins move in or out of the nucleus in minutes, after upstream signalling modifies their state of phosphorylation. Although such approaches are being used increasingly, there is no consensus on how to quantify the nuclear and cytoplasmic localisation of these proteins. Using budding yeast, we developed a convolutional neural network that quantifies nuclear localisation from fluorescence and, optionally, bright-field images. Focusing on the cellular response to changing glucose, we generated ground-truth data using strains with both a transcription factor and a nuclear protein tagged with fluorescent markers. We then showed that the neural network based approach outperformed seven published methods, particularly when predicting single-cell time series, which are key to determining how cells respond and adapt. Collectively, our results are conclusive — using machine learning to automatically determine the appropriate image processing consistently outperforms ad hoc methods. Adopting such methods promises to both improve the accuracy and, with transfer learning, the consistency of single-cell analyses.