High-throughput detection and tracking of cells and intracellular spots in mother machine experiments

Abstract

The analysis of bacteria at the single cell level is essential to characterize processes in which cellular heterogeneity plays an important role. BACMMAN (BACteria Mother Machine ANalysis) is a software allowing fast and reliable automated image analysis of high-throughput 2D or 3D time-series images from experiments using the "mother machine", a very popular microfluidic device allowing biological processes in bacteria to be investigated at the singlecell level. Here we describe how to use some of the BACMMAN features including i) segmentation and tracking of bacteria and intracellular fluorescent spots, ii) visualization and editing of the results iii) configuration of the image processing pipeline for different datasets, and iv) BACMMAN coupling to data analysis software for visualization and analysis of data subsets with specific properties. Among software specifically dedicated to the analysis of mother machine data, only BACMMAN allows segmentation and tracking of both bacteria and intracellular spots. For a single position, single channel with 1000 frames (2GB dataset) image processing takes about 6 min on a regular computer. Numerous implemented algorithms, easy configuration and high modularity ensure wide applicability of the BACMMAN software.

DeLTA: Automated cell segmentation, tracking, and lineage reconstruction using deep learning

“…For instance, the mother machine movie analysis software MoMa [16] reports tracking error rates of about 4%. Other proposed solutions do not explicitly report their accuracy and typically require some parameter tweaking to be adapted to new data, and often also require a posteriori correction [10,11,17,19,32]. In contrast, DeLTA can adapt to entirely new images from a different setup without the need for re-training if those new images are similar to those in the training set.…”

“…To create training sets, we began by using the Ilastik software [14] to segment time-lapse movies of cells in the mother machine chips. Note that other training set generation pipelines can be used, for example with other mother machine data analysis software [11,16,17,19], or if experimenters have pre-existing segmented sets. While our segmentation results with Ilastik were already fairly accurate (~1% error rate), we found that even after carefully designing an initial training set, the results generalized poorly to new datasets and usually required training sets to be generated or updated a posteriori with each new experiment.…”

“…The network was trained on images from three different multi-position time-lapse movies acquired in our lab, and on movies available from the literature [11,16,17,28,29]; we then evaluated on a third data set acquired in our lab weeks later as well as on data from the literature ( Table 1). To demonstrate generalizability of the approach, and to evaluate performance against other analysis software, we also used the time-lapse image datasets provided with the mother machine analysis software tools Molyso [11], MoMA [16], and BACMMAN [17] to evaluate the performance of DeLTA on completely new data by performing "leave-one-out" training and evaluation pairings. In these studies, one of those three different datasets was left out at the training step, and after training, performance was evaluated against the omitted dataset (S1 Table).…”

“…While a plethora of software suites have been developed for single-cell segmentation and tracking [12][13][14][15], including code specific to analysis of mother machine data [11,[16][17][18][19], the vast majority require manual inputs from the experimenter and are designed for a posteriori processing. The relatively recent breakthrough in biomedical image analysis brought by deep convolutional neural networks, and the U-Net [20] architecture in particular, has introduced an era of fast-paced developments in the field [21].…”

Mother machine image analysis with MM3

“…There are a number of mother machine-specific image analysis packages available [2][3][4][5][6][7][8][9] in addition to general cell image analysis packages which may be adapted to a mother machine workflow [10][11][12][13] . MM3 aims to be a complete and flexible solution to this problem, taking raw micrographs and producing readily graphable cell data.…”