cellmigRation

Importing TIFF files

In this vignette, we are going to analyze three images with the aim of illustrating the functions included in cellmigRation. The original TIFF files are available at the following URLs:

• https://www.data-pulse.com/projects/2020/cellmigRation/ctrl_001.tif

• https://www.data-pulse.com/projects/2020/cellmigRation/ctrl_002.tif

• https://www.data-pulse.com/projects/2020/cellmigRation/drug_001.tif

Note. TIFF files can be imported using the LoadTiff() function. This function includes a series of (optional) arguments to attach meta-information to a TIFF image, for example the experiment and condition arguments. Imported numeric images are stored as a trackedCells-class object.

Three sample trackedCells objects (imported from the corresponding TIFF files) are available as a list in the cellmigRation package (ThreeConditions object). These will be used for illustrating the functions of our package in this vignette.

# load data
data(ThreeConditions)

# An S4 trackedCells object
ThreeConditions[[1]]
#>  ~~~ An S4 trackedCells object ~~~
#> 
#>       + Num of images: 7
#>       + Optimized Params: No
#>       + Run w/ Custom Params: No
#>       + Cells Tracked: No
#>       + Stats Computed: No

Optimizing Tracking Params

This is an optional yet recommended step. Detecting fluorescent cells requires defining a series of parameters to maximize signal to noise ratio. Specifically,

• diameter: size corresponding to the largest diameter of a cell (expressed in pixels). Ideally, we want to set this parameter to a value large enough to capture all cells (even the large ones), but small enough to exclude aggregates or large background particles (artifacts, bubbles)

• lnoise: size corresponding to the smalles diameter of a cell (expressed in pixels). Ideally, we want to set this parameter to a value small enough to capture all cells (even the small ones), but large enough to exclude small background particles (artifacts, debris)

• threshold: signal level used as background threshold. Signal smaller than threshold is set to zero

If the values of these arguments are known, you can skip this step. Alternatively, if you want to test a specific range of these values, you can run OptimizeParams() manually specifying the ranges to be tested. By default, the function determines automatically a reasonable range of values to be tested for each argument based on the empirical distribution of signal and sizes of particles detected in the frame with median signal from a TIFF stack. This operation supports parallelization (recommended: parallelize by setting the threads argument to a value bigger than 1).

Note: the user may request to visualize a plot. The output plot shows how many cells were detected for each combination of parameter values. By default, the pick #1 is selected for the downstream steps.

Note 2: for larger datasets, the user may wish to set the threads argument below to a larger integer in order to benefit from paralellized operations. A theoretical upper bound to this argument would be the number of threads in your CPU—which you can check with parallel::detectCores()—, but it is considered good practice to leave at least one thread for other system operations.

# Optimize parameters using 1 core
x1 <- OptimizeParams(
    ThreeConditions$ctrl01, threads = 1, lnoise_range = c(5, 12),
    diameter_range = c(16, 22), threshold_range = c(5, 15, 30),
    verbose = FALSE, plot = TRUE)

Note 3: the getOptimizedParams() is a getter function to obtain the values of each optimized parameter.

# obtain optimized params
getOptimizedParams(x1)$auto_params
#> $lnoise
#> [1] 5
#> 
#> $diameter
#> [1] 16
#> 
#> $threshold
#> [1] 5

Tracking Cell Movements

The central step of Module 1 is tracking cell movements across all frames of a multi-stack image (where each stack was acquired at a different time). This operation is carried out via the CellTracker() function, which performs two tasks: i) identify all cells in each frame of the image; ii) map cells across all image frames, identify cell movements and return cell tracks. This operation supports parallelization. This function requires three parameters to be set: lnoise, diameter, and threshold. These parameters can be set manually or automatically:

• rely on the optimized params estimated using OptimizeParams()

• rely on the optimized params estimated for a different trackedCells object; using OptimizeParams(); see the import_optiParam_from argument

• the user can manually specify the parameter values; note that user-specified parameters will overwrite automatically-optimized values

Note 1: the user may request to visualize a plot for each frame being processed. The output plot shows cells that were detected for each combination of parameter values.

Note 2: it is possible to only include cells that were detected in at least a minimum number of frames by setting the min_frames_per_cell argument. If so, cells detected in a small number of frames will be removed from the output.

Note 3: the user may parallelize (recommended when possible) this step by setting the threads argument to a value bigger than 1.

# Track cell movements using optimized params
x1 <- CellTracker(
    tc_obj = x1, min_frames_per_cell = 3, threads = 1, verbose = TRUE)

# Track cell movements using params from a different object
x2 <- CellTracker(
    ThreeConditions$ctrl02, import_optiParam_from = x1,
    min_frames_per_cell = 3, threads = 1)

# Track cell movements using CUSTOM params, show plots
x3 <- CellTracker(
    tc_obj = ThreeConditions$drug01,
    lnoise = 5, diameter = 22, threshold = 6,
    threads = 1, maxDisp = 10,
    show_plots = TRUE)

It is possible to retrieve the output data.frame including information about cell movements (cell tracks) using the getTracks() getter function.

# Get tracks and show header
trk1 <- cellmigRation::getTracks(x1)
head(trk1) %>% kable() %>% kable_styling(bootstrap_options = 'striped')

Basic migration stats

For compatibility and portability reasons, Module 1 includes a function to compute the same basic metrics/stats as in the FastTracks Matlab software by Brian DuChez. This step is performed via the ComputeTracksStats() function. The results can be extracted from a trackedCells object via dedicated getter functions: getPopulationStats() and getCellsStats(). Note however that more advanced stats are computed using functions included in the second module of cellmigRation.

# Basic migration stats can be computed similar to the fastTracks software
x1 <- ComputeTracksStats(
    x1, time_between_frames = 10, resolution_pixel_per_micron = 1.24)
x2 <- ComputeTracksStats(
    x2, time_between_frames = 10, resolution_pixel_per_micron = 1.24)
x3 <- ComputeTracksStats(
    x3, time_between_frames = 10, resolution_pixel_per_micron = 1.24)

# Fetch population stats and attach a column with a sample label
stats.x1 <- cellmigRation::getCellsStats(x1) %>%
    mutate(Condition = "CTRL1")
stats.x2 <- cellmigRation::getCellsStats(x2) %>%
    mutate(Condition = "CTRL2")
stats.x3 <- cellmigRation::getCellsStats(x3) %>%
    mutate(Condition = "DRUG1")

stats.x1 %>%
    dplyr::select(
        c("Condition", "Cell_Number", "Speed", "Distance", "Frames")) %>%
    kable() %>% kable_styling(bootstrap_options = 'striped')

# Run a simple Speed test
sp.df <- rbind(
    stats.x1 %>% dplyr::select(c("Condition", "Speed")),
    stats.x2 %>% dplyr::select(c("Condition", "Speed")),
    stats.x3 %>% dplyr::select(c("Condition", "Speed"))
)

vp1 <- ggplot(sp.df, aes(x=Condition, y = Speed, fill = Condition)) +
    geom_violin(trim = FALSE) +
    scale_fill_manual(values = c("#b8e186", "#86e1b7", "#b54eb4")) +
    geom_boxplot(width = 0.12, fill = "#d9d9d9")

print(vp1)

# Run a t-test:
sp.lst <- with( sp.df, split(Speed, f = Condition))
t.test(sp.lst$CTRL1, sp.lst$DRUG1, paired = FALSE, var.equal = FALSE)
#> 
#>  Welch Two
#>  Sample
#>  t-test
#> 
#> data:  sp.lst$CTRL1 and sp.lst$DRUG1
#> t = 7.6628,
#> df =
#> 3.7144,
#> p-value =
#> 0.002093
#> alternative hypothesis: true difference in means is not equal to 0
#> 95 percent confidence interval:
#>  0.3876210 0.8499647
#> sample estimates:
#> mean of x 
#> 1.1582615 
#> mean of y 
#> 0.5394686

Basic Visualizations

Two basic visualization functions are included in Module 1, and allow visualization of cells detected in a frame of interest, and tracks originating at a frame of interest. These functions are included in Module 1 (and not Module 2) since they take a trackedCells-class object as input.

# Visualize cells in a frame of interest
cellmigRation::VisualizeStackCentroids(x1, stack = 1)

# Visualize tracks of cells originating at a frame of interest
par(mfrow = c(1, 3))
cellmigRation::visualizeCellTracks(x1, stack = 1, main = "tracks from CTRL1")
cellmigRation::visualizeCellTracks(x2, stack = 1, main = "tracks from CTRL2")
cellmigRation::visualizeCellTracks(x3, stack = 1, main = "tracks from DRUG1")

Aggregate Cell Tracks

Cell tracks from multiple TIFF images can be aggregated together. All tracks form the different experiments/images are returned in a large data.frame. A new unique ID is assigned to specifically identify each cell track from each image/experiment. Different trackedCells objects can be merged together based on the corresponding TIFF filename (default), or one of the meta-information included in the object(s).

Note 1: the data.frame returned by aggregateTrackedCells() has a structure that aligns to the output of the getTracks() function when the attach_meta argument is set to TRUE.

Note 2: the data.frame returned by aggregateTrackedCells() (or by getTracks() with the attach_meta argument set to TRUE) is the input of the CellMig() function, and is the first step of Module 2.

Note 3: it is recommended to aggregate experiments/tiff files corresponding to the same condition (as shown below: for example, all replicates of the control cells) However, it is also possible to mix and match multiple treatments/timepoints/conditions, and filter the desired tracks right before running the CellMig() step (not shown).

# aggregate tracks together
all.ctrl <- aggregateTrackedCells(x1, x2, meta_id_field = "tiff_file")

# Show header
all.ctrl[seq_len(10), seq_len(6)] %>%
    kable() %>% kable_styling(bootstrap_options = 'striped')

#>          tiff_file
#> condition CTRL01.tif
#>      CTRL         28
#>          tiff_file
#> condition CTRL02.tif
#>      CTRL         25

# Prepare second input of Module 2
all.drug <- getTracks(tc_obj = x3, attach_meta = TRUE)

Import and Pre-process Cell Tracks

The first step entails the generation of a CellMig-class object (S4 class) to store cell tracks data, and all output resulting from running Module 2 functions. After importing data into a CellMig-class object, tracks are processed according to the experiment type (random migration in a plate vs. scratch-wound healing assay).

Note 1: the arguments passed to the CellMig() function are:

• trajdata a data.frame, the output from the previous module

• expName a string, this is the name of the experiment

Note 1: the user is allowed to name the analysis; here we select a name that will be used as a prefix in the name of plots and tables.

Note 2: For Random Migration assays, the rmPreProcessing() function is used for preprocessing; if a Scratch Wound Healing Assay was performed, the wsaPreProcessing() function shall be used instead.

rmTD <- CellMig(trajdata = all.ctrl)
rmTD <- setExpName(rmTD, "Control")

# Preprocessing the data
rmTD <- rmPreProcessing(rmTD, PixelSize=1.24, TimeInterval=10, FrameN=3)
#> This dataset contains: 8 cell(s) in total
#> This dataset contains: 8 cell(s) with more than three steps in their tracks
#> The desired number of steps:  3 
#> The maximum number of steps:  7 
#> Only:  8  cells were selected 
#> All the tracks of the selected cells are adjusted to have only  3  steps

Plotting tracks (2D and 3D)

Multiple plotting functions allow the user to generate 2D or 3D charts and plots showing the movements of all cells, or part of the cells in the experiment.

# Plotting tracks (2D and 3D)
plotAllTracks(rmTD, Type="l", FixedField=FALSE, export=FALSE)
#> The plot contains 8 Cells

# Plotting the trajectory data of sample of cells (selected randomly)
# in one figure
plotSampleTracks(
    rmTD, Type="l", FixedField=FALSE, celNum=2, export = FALSE)
#> The plot contains the following cells:  
#> 2 4

3D Plots

The following functions are meant to be run in an interactive fashion:

• plot3DAllTracks(rmTD, VS=2, size=5)

• plot3DTracks(rmTD, cells=1:10, size = 8)

Deep Trajectory Analysis

The deep trajectory analysis includes a series of tools to examine the following metrics:

• Persistence and Speed: PerAndSpeed() function

• Directionality: DiRatio() function

• Mean Square Displacement: MSD() function

• Direction AutoCorrelation: DiAutoCor() function

• Velocity AutoCorrelation: VeAutoCor() function

These steps are meant to be run on larger datasets, including a larger number of cells. Here, we only show an example of how to run a DiRatio analysis, an MSD analysis and Velocity autocorrelation.

For more examples about Deep Trajectory Analysis, please visit: https://www.data-pulse.com/projects/2020/cellmigRation/cellmigRation_v01.html

Directionality Analysis. This analysis is performed via the DiRatio() function. Results are saved in a CSV file. Plots can be generated using the DiRatioPlot() function. Plots are saved in a newly created folder with the following extension: -DR_Results.

## Directionality
srmTD <- DiRatio(rmTD, export=TRUE)
#> Results are saved as:  Control-DRResultsTable.csv in your directory [use getwd()]
DiRatioPlot(srmTD, export=TRUE)
#> Plots are saved in a folder in your directory [use getwd()]

Controldirectionality

Mean Square Displacement. The MSD function automatically computes the mean square displacements across several sequential time intervals. MSD parameters are used to assess the area explored by cells over time. Usually, both the sLAG and ffLAG arguments are recommended to be set to 0.25 but since here we have only few frames per image, we will set it to 0.5.

rmTD<-MSD(object = rmTD, sLAG=0.5, ffLAG=0.5, export=TRUE)

Velocity AutoCorrelation. The VeAutoCor() function automatically computes the changes in both speed and direction across several sequantial time intervals. Usually the sLAG is recommended to be set to 0.25 but since here we have just few frames, we will set it to 0.5.

rmTD <- VeAutoCor(
    rmTD, TimeInterval=10, sLAG=0.5, sPLOT=TRUE,
    aPLOT=TRUE, export=FALSE)

Final Results

The FinRes() function automatically generates a data frame that contains all the results with or without the a correlation table.

rmTD <-FinRes(rmTD, ParCor=TRUE, export=FALSE)
#>  [1] "MSD (lag=1)"                             
#>  [2] "MSD slope"                               
#>  [3] "N-M best fit (Furth) [D]"                
#>  [4] "N-M best fit (Furth) [P]"                
#>  [5] "The significance of fitting D"           
#>  [6] "The significance of fitting P"           
#>  [7] "Velocity AutoCorrelation (lag=1)"        
#>  [8] "2nd normalized Velocity AutoCorrelation" 
#>  [9] "Intercept of VA quadratic model"         
#> [10] "Mean Velocity AutoCorrelation (all lags)"

Below, the first 5 columns of the output data.frame are shown.

Principal Component Analysis (PCA) and Cell Clustering

The CellMigPCA() function automatically generates Principal Component Analysis based on a set of parameters selected by the user. The CellMigPCAclust() function automatically generates clusters based on the Principal Component Analysis. This analysis is supposed to be run in an interactive session via the CellMigPCA() function.