Advanced Usage

ELLA has been tested on high-resolution spatial transcriptomics datasets across various platforms and technologies. It comes with a set of default argument values that can be customized as needed. The usage of these customizable arguments is introduced in this page.

The full list of customizable arguments and their default choices and functions are listed in the table below

Args	Type	Default	Function
dataset	str	‘untitled’	Name of the dataset, help to distinguish multiple runs
beta_kernel_param_list	list of lists	22 lists	Shape parameters of the 22 beta kernel functions in NHPP model fitting
adam_learning_rate_max	float	1e-2	Max initial learning rate of Adam
adam_learning_rate_min	float	1e-3	Min initial learning rate of Adam
adam_learning_rate_adjust	float	1e7	Adam LR = loglikelihood value under the null divided by 1e-7
adam_delta_loss_max	float	1e-2	Max delta loss for Adam early stopping
adam_delta_loss_min	float	1e-5	Min delta loss for Adam early stopping
adam_delta_loss_adjust	float	1e8	Delta loss = loglikelihood value under the null divided by 1e-8
adam_niter_loss_unchange	int	20	Adam stops if loss decrease < delta loss for 20 iterations
max_iter	int	5e3	Max number of interations in Adam
min_iter	int	1e2	Min bumber of interations in Adam
max_ntanbin	int	25	Number of bins for computing relative positions
ri_clamp_min	float	1e-2	Min relative position
ri_clamp_max	float	1.0	Max relative position
hpp_solution	str	‘numercial’	Use numercial solution for HPP
lam_filter	float	0.0	Exclude significant genes with max(lam)-min(lam) <= 0.0

The default values can be costomized while instantiating the class, for example

ella_demo = EG_analysis(dataset='Demo', max_iter=3000)