fit_periodic_loess: Fit periodic loess line with circular predictor

In hansenlab/tricycle: tricycle: Transferable Representation and Inference of cell cycle

Fit periodic loess line with circular predictor

Description

The function will fit a loess line using cell cycle position and other variables, such as expression levels of a gene or log-transformed totalUMIs numbers. The circular nature of cell cycle position is taken into account by making 3 copies inside the function. For convenience, the function will also return a scatter plot with fitted line if needed.

Arguments

theta.v	The cell cycle position - a numeric vector with range between 0 to 2pi.
y	The response variable - a numeric vector with the same length as theta.v.
span	The parameter α which controls the degree of smoothing. See loess. Default: 0.3
length.out	The number of data points on the fitted lines to be output in the prediction data.frame. Default: 200
plot	If TRUE, a ggplot scatter plot will be included in the output list. The figure will plot y ~ theta.v with points and the fitted loess line. Default: FALSE
fig.title	The title of the figure. Default: NULL
point.size	The size of the point in scatter plot used by geom_scattermore. Default: 2.1
point.alpha	The alpha value (transparency) of the point in scatter plot used by geom_scattermore. Default: 0.6
line.size	The size of the fitted line, used by geom_path. Default: 0.8
line.alpha	The alpha value (transparency) of the fitted line, used by geom_path. Default: 0.8
color.vars	Optional. A vector of categorical variable of the same length of theta.v, and it will be used to color points in figure. Default: NULL
color.name	The name of the color variables. Used as the name for legend. Default: NULL
x_lab	Title of x-axis. Default: "θ"
y_lab	Title of y-axis. Default: "y"
hue.colors	The string vector gives custom colors. If not given, the default scale_color_discrete will be used. Default: NULL
...	Other arguments input to loess.

Details

This function fit a normal loess line, but take the circularity of cell cycle position into account by making theta.v 3 periods (c(theta.v - 2 * pi, theta.v, theta.v + 2 * pi)) and repeating y 3 times. Only the fitted values corresponding to original theta.v will be returned. For convenience, the function will also return a scatter plot with fitted line if needed. Or user can use pred.df to visualize the loess line themselves.

Value

A list with the following elements:

• fitted - The fitted vaues on the loess line. A vector of the length of y.

• residual - The residual values from the fitted loess line, i.e. y - y.fit. A vector of the length of y.

• pred.df - The prediction data.frame by uniformly sampling theta from 0 - 2*pi. Names of variables: x and y. The number of rows equals to length.out.

• loess.o - The fitted loess object.

• rsquared - The coefficient of determination R2. Calculated as 1 - residual sum of squares / the total sum of squares.

• fig - When plot is TRUE, a ggplot scatter plot object will be returned with other items.

Author(s)

Shijie C. Zheng

See Also

estimate_cycle_position, for inferring cell cycle position.

Examples

data(neurosphere_example, package = "tricycle")
neurosphere_example <- estimate_cycle_position(neurosphere_example)
top2a.idx <- which(rowData(neurosphere_example)$Gene == "Top2a")
fit.l <- fit_periodic_loess(neurosphere_example$tricyclePosition,
 assay(neurosphere_example, "logcounts")[top2a.idx, ], plot = TRUE)
fit.l$fig

hansenlab/tricycle documentation built on March 19, 2022, 7:24 p.m.