Package 'retrofit' reference manual

Title:	RETROFIT: Reference-free deconvolution of cell mixtures in spatial transcriptomics
Description:	RETROFIT is a Bayesian non-negative matrix factorization framework to decompose cell type mixtures in ST data without using external single-cell expression references. RETROFIT outperforms existing reference-based methods in estimating cell type proportions and reconstructing gene expressions in simulations with varying spot size and sample heterogeneity, irrespective of the quality or availability of the single-cell reference. RETROFIT recapitulates known cell-type localization patterns in a Slide-seq dataset of mouse cerebellum without using any single-cell data.
Authors:	Adam Park [aut, cre], Roopali Singh [aut] , Xiang Zhu [aut] , Qunhua Li [aut]
Maintainer:	Adam Park <[email protected]>
License:	GPL-3
Version:	1.7.0
Built:	2024-11-18 04:09:39 UTC
Source:	https://github.com/bioc/retrofit

RETROFIT matching algorithm

Description

Match cell types based on correlations with reference. decomp_w between matching algorithm description

Usage

annotateWithCorrelations(sc_ref, K, decomp_w, decomp_h)
annotateWithCorrelations(sc_ref, K, decomp_w, decomp_h)

Arguments

`sc_ref`	A Matrix or Array with two dimensions (GeneExpressions, Cell types).
`K`	integer: The number of cell types to be selected
`decomp_w`	Matrix(GeneExpressions, Components): Decomposed w matrix
`decomp_h`	Matrix(Components, Spots): Decomposed h matrix

Value

A list of selected components, cells, and correlations

w: Filtered 2d array with GeneExpressions, Cell types
h: Filtered2d array with Cell types, Spots
ranked_cells: The list of cell names
ranked_correlations: The list of correlations

Examples

data("testSimulationData")
K            = 10
sc_ref       = testSimulationData$sc_ref
W            = testSimulationData$decompose$w
H            = testSimulationData$decompose$h

result       = retrofit::annotateWithCorrelations(sc_ref=sc_ref, K=K, 
                                                 decomp_w=W, decomp_h=H)
H_annotated  = result$h                                              
W_annotated  = result$w
ranked_cells = result$ranked_cells
data("testSimulationData")
K            = 10
sc_ref       = testSimulationData$sc_ref
W            = testSimulationData$decompose$w
H            = testSimulationData$decompose$h

result       = retrofit::annotateWithCorrelations(sc_ref=sc_ref, K=K, 
                                                 decomp_w=W, decomp_h=H)
H_annotated  = result$h                                              
W_annotated  = result$w
ranked_cells = result$ranked_cells

RETROFIT matching algorithm

Description

Match cell types based on correlations with reference. decomp_w between matching algorithm description

Usage

annotateWithMarkers(marker_ref, K, decomp_w, decomp_h)
annotateWithMarkers(marker_ref, K, decomp_w, decomp_h)

Arguments

`marker_ref`	Key-value list: A dictionary of key: cell type, value: GeneExpression list
`K`	integer: The number of cell types to be selected
`decomp_w`	Matrix(GeneExpressions, Components): Decomposed w matrix
`decomp_h`	Matrix(Components, Spots): Decomposed h matrix

Value

A list of

Examples

data("testSimulationData")
K            = 10
marker_ref   = testSimulationData$marker_ref
W            = testSimulationData$decompose$w
H            = testSimulationData$decompose$h

result       = retrofit::annotateWithMarkers(marker_ref=marker_ref, K=K, 
                                            decomp_w=W, decomp_h=H)
H_annotated  = result$h                                              
W_annotated  = result$w
ranked_cells = result$ranked_cells
data("testSimulationData")
K            = 10
marker_ref   = testSimulationData$marker_ref
W            = testSimulationData$decompose$w
H            = testSimulationData$decompose$h

result       = retrofit::annotateWithMarkers(marker_ref=marker_ref, K=K, 
                                            decomp_w=W, decomp_h=H)
H_annotated  = result$h                                              
W_annotated  = result$w
ranked_cells = result$ranked_cells

RETROFIT decomposition algorithm

Description

Receiving the input with 2d spatial transcriptomics matrix, the function returns factorized W, H, Theta. This function fulfills Structured Stochastic Variational Inference Algorithm for RETROFIT. Since exact Bayesian inference is infeasible and considering the large number of spots and genes, variational inference was adopted to approximately estimate the parameters in performant manner.

Usage

decompose(
  x,
  L = 16,
  iterations = 4000,
  init_param = NULL,
  lambda = 0.01,
  kappa = 0.5,
  verbose = FALSE
)
decompose(
  x,
  L = 16,
  iterations = 4000,
  init_param = NULL,
  lambda = 0.01,
  kappa = 0.5,
  verbose = FALSE
)

Arguments

`x`	matrix or array with dimension (GeneExpressions, Spots). This is the main spatial transciptomics data.
`L`	integer (default:16) The number of components to be decomposed
`iterations`	integer (default:4000) The number of maximum iterations to be executed
`init_param`	list Vatirational initial parameters
`lambda`	double (default:0.01) Background expression profile control
`kappa`	double (default:0.5) Learning rate factor
`verbose`	boolean (default:FALSE)

Details

init_param specification

alpha_w_0 double (default:0.05)
beta_w_0 double (default:0.0001)
alpha_h_0 double (default:0.2)
beta_h_0 double (default:0.2)
alpha_th_0 double (default:1.25)
beta_th_0 double (default:10)
alpha_th_k array (default:array with dim c(K))
beta_th_k array (default:array with dim c(K)),
alpha_w_gk array (default:array with dim c(G,K)),
beta_w_gk array (default:array with dim c(G,K)),
alpha_h_ks array (default:array with dim c(K,S)),
beta_h_ks array (default:array with dim c(K,S))

Value

A list of decomposed vectors that contains

w: 2d array with GeneExpressions, Components
h: 2d array with Components, Spots
th: an array with Components
durations: (verbose) durations vector (unit: second)
relative_error:(verbose) errors with pre-defined norm vector

Examples

data("testSimulationData")
x   = testSimulationData$extra5_x
res = retrofit::decompose(x, L=16, iterations=10, verbose=TRUE)
W   = res$w
H   = res$h
TH  = res$th
data("testSimulationData")
x   = testSimulationData$extra5_x
res = retrofit::decompose(x, L=16, iterations=10, verbose=TRUE)
W   = res$w
H   = res$h
TH  = res$th

RETROFIT

Description

The main algorithm

Usage

retrofit(
  x,
  sc_ref = NULL,
  marker_ref = NULL,
  L = 16,
  K = 8,
  iterations = 4000,
  init_param = NULL,
  lambda = 0.01,
  kappa = 0.5,
  verbose = FALSE
)
retrofit(
  x,
  sc_ref = NULL,
  marker_ref = NULL,
  L = 16,
  K = 8,
  iterations = 4000,
  init_param = NULL,
  lambda = 0.01,
  kappa = 0.5,
  verbose = FALSE
)

Arguments

`x`	A matrix or array with dimension (GeneExpressions, Spots). This is the main spatial transciptomics data.
`sc_ref`	A matrix or array with two dimensions (GeneExpressions, Cell types).
`marker_ref`	A list with (keys, values) = (cell types, an array of genes).
`L`	integer (default:16) The number of components to be decomposed
`K`	integer: The number of cell types to be selected
`iterations`	integer (default:4000) The number of maximum iterations to be executed
`init_param`	list Vatirational initial parameters
`lambda`	double (default:0.01) Background expression profile control
`kappa`	double (default:0.5) Learning rate factor
`verbose`	boolean (default:FALSE)

Value

A list of decomposed vectors that contains

decompose:
- w: Decomposed 2d array with GeneExpressions, Components
- h: Decomposed 2d array with Components, Spots
- th: 1d array with Components
annotateWithCorrelations:
- w: Filtered 2d array with GeneExpressions, Cell types
- h: Filtered2d array with Cell types, Spots
annotateWithMarkers:
- w: Filtered 2d array with GeneExpressions, Cell types
- h: Filtered2d array with Cell types, Spots

Examples

data("testSimulationData")
iterations  = 10
L           = 16
K           = 8
x           = testSimulationData$extra5_x
sc_ref      = testSimulationData$sc_ref

res         = retrofit::retrofit(x, sc_ref=sc_ref, L=L, K=K, iterations=iterations)
W           = res$decompose$w
W_annotated = res$annotateWithCorrelations$w
ranked_cells= res$annotateWithCorrelations$ranked_cells
data("testSimulationData")
iterations  = 10
L           = 16
K           = 8
x           = testSimulationData$extra5_x
sc_ref      = testSimulationData$sc_ref

res         = retrofit::retrofit(x, sc_ref=sc_ref, L=L, K=K, iterations=iterations)
W           = res$decompose$w
W_annotated = res$annotateWithCorrelations$w
ranked_cells= res$annotateWithCorrelations$ranked_cells

simulation data

Description

A dataset with input and output of retrofit functions for reproducibility tests.

Usage

data(testSimulationData)
data(testSimulationData)

Format

Includes input x, references and results with large iterations

Details

testSimulationData

colon vignette

Description

A dataset supporting the colon vignette process

Usage

data(vignetteColonData)
data(vignetteColonData)

Format

Includes colon scenario x, references, a large iterations results.

Details

vignetteColonData

simulation vignette

Description

A dataset supporting the simulation vignette process

Usage

data(vignetteSimulationData)
data(vignetteSimulationData)

Format

Includes n10m3 scenario x, references, a large iterations results.

Details

vignetteSimulationData

Package 'retrofit'

Help Index

RETROFIT matching algorithm

Description

Usage

Arguments

Value

See Also

Examples

RETROFIT matching algorithm

Description

Usage

Arguments

Value

See Also

Examples

RETROFIT decomposition algorithm

Description

Usage

Arguments

Details

Value

See Also

Examples

RETROFIT

Description

Usage

Arguments

Value

See Also

Examples

simulation data

Description

Usage

Format

Details

colon vignette

Description

Usage

Format

Details

simulation vignette

Description

Usage

Format

Details