Package 'inferCSN'

Inferring Cell-Specific Gene Regulatory Network

Description

Inferring cell-type specific gene regulatory network using sparse regression model from single-cell transcriptomic data.

Author(s)

Meng Xu (Maintainer), [email protected]

Source

https://mengxu98.github.io/inferCSN/

See Also

Useful links:

• https://mengxu98.github.io/inferCSN/

• Report bugs at https://github.com/mengxu98/inferCSN/issues

---

Example ground truth data

Description

Ground-truth edges for the bundled synthetic example.

---

Example matrix data

Description

Expression matrix for the bundled synthetic example.

---

Example meta data

Description

Cell metadata for the bundled synthetic example.

---

Filter and sort a network matrix

Description

Filter and sort a network matrix

Usage

filter_sort_matrix(network_matrix, regulators = NULL, targets = NULL)

Arguments

network_matrix	Network weight matrix.
regulators, targets	Nodes to include.

Value

A filtered and sorted matrix.

---

Fit greedy L0 regression

Description

Fits a standardized sparse linear model by BIC-decreasing support updates.

Usage

fit_greedy_l0(
  x,
  y,
  max_support_size = NULL,
  min_improvement = 1e-10,
  verbose = TRUE
)

Arguments

x	Numeric predictor matrix with observations in rows.
y	Numeric response vector.
max_support_size	Optional positive integer support limit.
min_improvement	Minimum RSS improvement considered.
verbose	Logical verbosity flag.

Value

A list with 'model', 'metrics', and 'coefficients'.

---

Fit batched greedy L0 regressions

Description

Fits multiple responses from shared predictor cross-products.

Usage

fit_greedy_l0_batch(
  gram,
  xty,
  response_ss,
  candidates,
  n_obs,
  max_support_size = NULL,
  min_improvement = 1e-10
)

Arguments

gram	Numeric square predictor cross-product matrix.
xty	Numeric predictor-by-response cross-product matrix.
response_ss	Numeric response sums of squares.
candidates	List of one-based predictor indices, one element per response.
n_obs	Number of observations used for the cross-products.
max_support_size	Optional positive integer support limit.
min_improvement	Minimum RSS reduction proposed during forward search.

Value

Batched coefficients, deletion evidence, and fit summaries.

---

inferring cell-type specific gene regulatory network

Description

inferring cell-type specific gene regulatory network

Usage

inferCSN(
  object,
  pseudotime = NULL,
  regulators = NULL,
  targets = NULL,
  max_support_size = NULL,
  lag_fraction = 0.05,
  lag_steps = NULL,
  cores = 1,
  verbose = TRUE,
  method = c("greedy_l0", "L0", "L0L1", "L0L2"),
  ...
)

## S4 method for signature 'matrix'
inferCSN(
  object,
  pseudotime = NULL,
  regulators = NULL,
  targets = NULL,
  max_support_size = NULL,
  lag_fraction = 0.05,
  lag_steps = NULL,
  cores = 1,
  verbose = TRUE,
  method = c("greedy_l0", "L0", "L0L1", "L0L2"),
  ...
)

## S4 method for signature 'sparseMatrix'
inferCSN(
  object,
  pseudotime = NULL,
  regulators = NULL,
  targets = NULL,
  max_support_size = NULL,
  lag_fraction = 0.05,
  lag_steps = NULL,
  cores = 1,
  verbose = TRUE,
  method = c("greedy_l0", "L0", "L0L1", "L0L2"),
  ...
)

Arguments

object	Numeric expression matrix with cells in rows and genes in columns.
pseudotime	Optional pseudotime vector or branch matrix for either method. Regulators in earlier states predict targets in later states; tied states are averaged and shared branch transitions are counted once.
regulators, targets	Optional gene subsets.
max_support_size	Optional support-size limit.
lag_fraction	Fractional lag used when lag_steps is NULL.
lag_steps	Optional integer lag. L0Learn centers and scales lagged expression within each branch and defaults to fitting without an intercept.
cores	Number of inference workers.
verbose	Whether to report progress.
method	greedy_l0 (default), or L0Learn with the L0, L0L1, or L0L2 penalty.
...	Arguments passed to the method.

Value

A data frame containing exactly regulator, target, and weight.

Examples

data(example_matrix)
data(example_meta_data)
network_table <- inferCSN(
  example_matrix,
  pseudotime = example_meta_data$pseudotime
)
head(network_table)

inferCSN(
  example_matrix,
  regulators = c("g1", "g2"),
  targets = c("g3", "g4")
)

---

The logo of inferCSN

Description

The inferCSN logo, using ASCII or Unicode characters Use cli::ansi_strip to get rid of the colors.

Usage

infercsn_logo(unicode = cli::is_utf8_output())

Arguments

unicode

Unicode symbols on UTF-8 platforms. Default is cli::is_utf8_output.

Value

A character vector.

References

https://github.com/tidyverse/tidyverse/blob/main/R/logo.R

Examples

infercsn_logo()

---

Format a network table

Description

Format a network table

Usage

network_format(
  network_table,
  regulators = NULL,
  targets = NULL,
  abs_weight = TRUE
)

Arguments

network_table	Network edge table.
regulators, targets	Nodes to include.
abs_weight	Whether to use absolute weights and add interaction signs.

Value

A formatted network edge table.

---

Print logo

Description

Print logo

Usage

## S3 method for class 'infercsn_logo'
print(x, ...)

Arguments

x	Input information.
...	Other parameters.

Value

Print the ASCII logo

---

Select features with significant trends

Description

Select features with significant trends

Usage

select_trend_features(
  x = NULL,
  pseudotime = NULL,
  method = c("pretsa", "gam"),
  padjust_threshold = 0.05,
  n_candidates = NULL,
  statistics = NULL,
  ...
)

Arguments

x	Numeric matrix with observations in rows and features in columns, or NULL when supplying statistics.
pseudotime	Numeric coordinate per observation.
method	Fitting method: "pretsa" or "gam".
padjust_threshold	Strict adjusted P-value cutoff between zero and one.
n_candidates	Maximum number of selected features, or NULL for no cap.
statistics	Precomputed table with feature row names and padjust (or pvalue if padjust is absent).
...	Numerical options passed to thisutils::fit_trends().

Value

A list with features, statistics, and fit (NULL for precomputed input). Nonfinite P-values are excluded. Input order is retained unless the cap is exceeded; then the smallest P-values are selected with stable ties.

Examples

t <- seq(0, 1, length.out = 40)
x <- cbind(a = sin(t * 5), b = cos(t * 3))
select_trend_features(x, t)$features

---

Construct network for single target gene

Description

Construct network for single target gene

Usage

single_network(
  matrix,
  regulators,
  target,
  pseudotime = NULL,
  max_support_size = NULL,
  lag_fraction = 0.05,
  lag_steps = NULL,
  cores = 1,
  verbose = TRUE,
  method = c("greedy_l0", "L0", "L0L1", "L0L2"),
  ...
)

Arguments

matrix	An expression matrix.
regulators	Candidate regulator genes.
target	The target gene.
pseudotime	Optional pseudotime vector or branch matrix passed to [inferCSN()].
max_support_size	Optional support-size cap passed to [inferCSN()].
lag_fraction	Fractional state lag passed to [inferCSN()].
lag_steps	Optional integer state lag passed to [inferCSN()].
cores	Number of inference workers.
verbose	Whether to report progress.
method	greedy_l0 (default), or L0Learn with the L0, L0L1, or L0L2 penalty.
...	Arguments passed to the method.

Value

A data frame with regulator, target, and weight columns. Greedy-L0 returns selected edges. L0Learn retains the original per-regulator coefficients, including zeros; [inferCSN()] removes zero weights from the complete network.

Examples

data(example_matrix)
head(
  single_network(
    example_matrix,
    regulators = colnames(example_matrix),
    target = "g1"
  )
)
single_network(
  example_matrix,
  regulators = c("g1", "g2", "g3"),
  target = "g1"
)

---