Compute Phi (ϕ) from a PI matrix — computePHI • scIso

This function estimates the dispersion parameter $\phi$ of a Beta distribution for each isoform (row) in a matrix of PI (Percent Inclusion) values. The parameter $\phi$ reflects the intercellular variability of isoform usage across cells.

Usage

computePHI(pi_mat)

Arguments

pi_mat: A numeric or sparse matrix of PI values (isoforms × cells).

Value

A named numeric vector of $\phi$ values, one per isoform (row of pi_mat). Rows with insufficient data or invalid parameters return NA.

Details

For each isoform, $\phi$ is estimated from the sample mean ($m$) and variance ($v$) of PI values across cells, using the method of moments: $$\phi = 1 / (\alpha + \beta + 1)$$ where $\alpha = m((1 - m)/v - 1)$ and $\beta = (1 - m)((1 - m)/v - 1)$.

Isoforms with fewer than three valid observations or with degenerate mean or variance values return NA.

Examples

set.seed(123)
pi_mat <- matrix(runif(20, 0.1, 0.9), nrow = 5,
                 dimnames = list(paste0("Iso", 1:5), paste0("Cell", 1:4)))
phi_vec <- computePHI(pi_mat)
phi_vec
#>      Iso1      Iso2      Iso3      Iso4      Iso5 
#> 0.5245752 0.1284286 0.3442660 0.1357083 0.4487447