1 About

This guide describes lute’s generics, methods, and classes for algorithms, including deconvolution and marker selection algorithms. This may be useful to algorithm developers and researchers interested in conducting systematic algorithm benchmarks.

1.1 Background

The class structure used by lute is based on the bluster R/Bioconductor package. It expands on that class structure by defining a hierarchy.

1.2 Motivation

Many algorithms are maintained and versioned in GitHub or Zenodo rather than a routinely versioned repository such as Bioconductor or CRAN. This can prove an obstacle when tracing package development and attempting comprehensive benchmarks, as software that is not actively maintained can become deprecated over time, and not all software will use compatible dependency versions (Maden et al. (2023)).

lute classes can helo with this to (1.) encourage use of common Bioconductor object classes (e.g. SummarizedExperiment, SingleCellExperiment, DelayedArray, etc.) and (2.) to use more standard inputs and outputs to encourage code reuse, discourage duplicated efforts, and enable more rapid and exhaustive benchmarks.

2 Classes

In a general sense, the class hierarchy is a wrapper allowing access to many algorithms using a single function and shared methods. However, it is possible to share data reformatting and preprocessing tasks, making the hierarchy more effectively similar to a workflow.

2.1 typemarkerParam

Topmost parameter class for cell type gene markers. This is used to manage the marker IDs.

2.2 deconvolutionParam

This is the parent class for all deconvolution algorithm param objects. The deconvolutionParam class is minimal, and simply defines slots for bulkExpression, or a matrix of bulk expression data, and returnInfo, a logical value indicating whether the default algorithm output will be stored and returned with standard output from running the deconvolution() method on a valid algorithm param object.

2.2.1 referencebasedParam

As shown in the class hierarchy diagram (above), referencebasedParam is a parent subclass inheriting attributes from deconvolutionParam. It is meant to contain and manage all tasks shared by reference-based deconvolution algorithms, or algorithms that utilize a cell type summary dataset. This is to be distinguished from reference-free algorithms.

This param class adds slots for referenceExpression, the cell type reference data, and cellScaleFactors, an optional vector of cell type size factors used to transform the reference.

2.2.2 independentbulkParam

This class is a subset of referencebasedParam algorithms specifying explicit samples used separately, such as for discrete training and test stages.

This param class adds a slot called bulkExpressionIndependent, which is for a dataset of bulk samples independent from samples specified in the bulkExpression slot.

2.3 Helper functions

lute provides a number of helper functions used to make the algorithm classes work. These include the parent classes and subclasses, and several functions to convert between object classes. These helper functions may be useful to developers. The following table indicates the functions and a short summary of what they do.

function_name description
referenceFromSingleCellExperiment() Makes the Z cell atlas reference from a SingleCellExperiment.
eset_to_sce() Convert ExpressionSet to SingleCellExperiment.
sce_to_eset() Convert SingleCellExperiment to ExpressionSet
se_to_eset() Convert SummarizedExperiment to ExpressionSet.
get_eset_from_matrix() Makes an ExpressionSet from a matrix.
parseDeconvolutionPredictionsResults() Gets formatted predicted cell type proportions table from deconvolution results list.
show() Method to inspect and summarize param object contents.
deconvolution() Method to perform deconvolution with a param object.
typemarkers() Method to get cell type markers with a param object.
deconvolutionParam() Defines the principal parent class for all deconvolution method parameters.
referencebasedParam() Class and methods for managing reference-based deconvolution methods.
independentbulkParam() Class and methods for managing methods requiring independent bulk samples.
typemarkersParam() Main constructor for class to manage mappings to the typemarkers() generic.

3 Algorithms

3.1 findMarkers

The param class findmarkersParam is defined for the function findMarkers() from scran (see ?findmarkersParam). This is a function to identify cell type marker genes from a single-cell or single-nucleus expression dataset.

The findmarkersParam class is organized under its parent classes as typemarkersParam->findMarkersParam. It includes the typemarkers() method for the identification of marker genes, and show() for inspecting the param contents.

The following images annotate the constructor function and the typemarkers() generic defined for the findmarkersParam class.

3.2 NNLS

The param class nnlsParam is defined for the function nnls from the nnls R/CRAN package (see ?nnlsParam). Non-negative least squares (NNLS) is commonly used for deconvolution.

The nnlsParam class is organized under its parent classes as deconvolutionParam->referencebasedParam->nnlsParam. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting the param contents.

The following images annotate the constructor function and the deconvolution() generic defined for the nnlsParam class.

3.3 Bisque

The param class bisqueParam is defined for the function ReferenceBasedDeconvolution from the BisqueRNA R/Bioconductor package (see ?bisqueParam). The Bisque algorithm adjusts on assay-specific biases arising between the bulk and single-cell or single-nucleus platforms used to generate expression datasets for deconvolution.

The bisqueParam class is organized under its parent classes as deconvolutionParam->referencebasedParam->independentbulkParam->bisqueParam. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting the param contents.

The following images annotate the constructor function and the deconvolution() generic defined for the bisqueParam class.

4 Extensions

We demonstrated the extensibility and flexibility of lute’s generic, method, and class system by extending support for additional algorithms beyond the 3 described above.

These algorithms can be used by sourcing the provided R/GitHub packages which pair the classes and functions with YML files for easier dependency management.

4.1 meanRatios

The param class meanratiosParam is defined for the function get_mean_ratios2() from the DeconvoBuddies R/GitHub package at LieberInstitute/DeconvoBuddies. This function uses the mean of cell type summary ratios to rank and select for top marker genes.

The meanratiosParam class is organized under its parent classes as typemarkersParam->meanratiosParam. It includes the typemarkers() generic for the identification of marker genes, and the show() method for inspecting the param contents.

meanratiosParam is available from GitHub at metamaden/meanratiosParam.

4.2 DeconRNASeq

deconvolutionParam->referencebasedParam->deconrnaseqParam.

The param class deconrnaseqParam is defined for the function DeconRNASeq (see ?deconrnaseqParam) from the DeconRNASeq R/Bioconductor package (link). The DeconRNASeq algorithm uses weighted averaged expression between types to predicted cell type amounts more accurately for heterogeneous tissues (Gong and Szustakowski (2013)).

The deconrnaseqParam class is organized under its parent classes as deconvolutionParam->referencebasedParam->deconrnaseqParam. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting the param contents.

The deconrnaseqParam class is available from GitHub at metamaden/deconrnaseqParam

4.3 EPIC

The param class epicParam is defined for the function EPIC from the EPIC R/GitHub package (see ?epicParam). The EPIC algorithm was developed in blood samples and incorporates cell size mRNA abundance (i.e. cell size) and variance normalizations (Racle and Gfeller (2020)).

The epicParam class is organized under its parent classes as deconvolutionParam->referencebasedParam->epicParam. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting the param contents.

The epicParam class is available from GitHub at metamaden/epicParam

4.4 MuSiC

The param class musicParam is defined for the function ReferenceBasedDeconvolution from the MuSiC R/GitHub package (see ?musicParam). The MuSiC algorithm adjusts on between-source variances for reference data from multiple sources (Wang et al. (2019)).

The musicParam class is organized under its parent classes as deconvolutionParam->referencebasedParam->musicParam. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting the param contents.

The musicParam class is available from GitHub at metamaden/musicParam

4.5 MuSiC2

The param class music2Param is defined for 2 implementations of the MuSiC2 algorithm from the MuSiC and MuSiC2 R/GitHub packages, respectively (see ?music2Param). The MuSiC2 algorithm pairs the features of the MuSiC algorithm with an additional filter for marker genes differentially expressed between cases and controls in the bulk and expression datasets (Fan et al. (2022)).

The music2Param class is organized under its parent classes as deconvolutionParam->referencebasedParam->independentbulkParam->music2Param. It includes the deconvolution() generic for cell type deconvolution, and the show() method for inspecting param contents.

The music2Param class is available from GitHub at metamaden/music2Param

5 Conclusions

This vignette showed how lute’s classes and methods are extensible and modular, and can encourage further development with standard algorithm I/O and object class management. First, we described lute’s algorithm class hierarchy, including how its parent classes and subclasses manage common tasks shared among algorithms and a table of functions developers may find useful. Further, we showed the annotated class and generic functions for algorithms supported by lute out of the box. Finally, we detail additional algorithms supported by R/GitHub packages that may be individually installed.

6 Session info

## R version 4.4.0 beta (2024-04-15 r86425)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 22.04.4 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.19-bioc/R/lib/libRblas.so 
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## time zone: America/New_York
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats4    stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
##  [1] lute_1.0.0                  SingleCellExperiment_1.26.0
##  [3] SummarizedExperiment_1.34.0 Biobase_2.64.0             
##  [5] GenomicRanges_1.56.0        GenomeInfoDb_1.40.0        
##  [7] IRanges_2.38.0              S4Vectors_0.42.0           
##  [9] BiocGenerics_0.50.0         MatrixGenerics_1.16.0      
## [11] matrixStats_1.3.0           BiocStyle_2.32.0           
## 
## loaded via a namespace (and not attached):
##  [1] xfun_0.43                 bslib_0.7.0              
##  [3] lattice_0.22-6            vctrs_0.6.5              
##  [5] tools_4.4.0               generics_0.1.3           
##  [7] parallel_4.4.0            tibble_3.2.1             
##  [9] fansi_1.0.6               cluster_2.1.6            
## [11] BiocNeighbors_1.22.0      pkgconfig_2.0.3          
## [13] Matrix_1.7-0              dqrng_0.3.2              
## [15] sparseMatrixStats_1.16.0  lifecycle_1.0.4          
## [17] GenomeInfoDbData_1.2.12   compiler_4.4.0           
## [19] statmod_1.5.0             bluster_1.14.0           
## [21] codetools_0.2-20          htmltools_0.5.8.1        
## [23] sass_0.4.9                yaml_2.3.8               
## [25] pillar_1.9.0              crayon_1.5.2             
## [27] jquerylib_0.1.4           BiocParallel_1.38.0      
## [29] limma_3.60.0              DelayedArray_0.30.0      
## [31] cachem_1.0.8              abind_1.4-5              
## [33] metapod_1.12.0            locfit_1.5-9.9           
## [35] tidyselect_1.2.1          rsvd_1.0.5               
## [37] digest_0.6.35             BiocSingular_1.20.0      
## [39] dplyr_1.1.4               bookdown_0.39            
## [41] fastmap_1.1.1             grid_4.4.0               
## [43] cli_3.6.2                 SparseArray_1.4.0        
## [45] magrittr_2.0.3            S4Arrays_1.4.0           
## [47] utf8_1.2.4                edgeR_4.2.0              
## [49] DelayedMatrixStats_1.26.0 UCSC.utils_1.0.0         
## [51] rmarkdown_2.26            XVector_0.44.0           
## [53] httr_1.4.7                igraph_2.0.3             
## [55] scran_1.32.0              ScaledMatrix_1.12.0      
## [57] beachmat_2.20.0           evaluate_0.23            
## [59] knitr_1.46                irlba_2.3.5.1            
## [61] rlang_1.1.3               Rcpp_1.0.12              
## [63] scuttle_1.14.0            glue_1.7.0               
## [65] BiocManager_1.30.22       jsonlite_1.8.8           
## [67] R6_2.5.1                  zlibbioc_1.50.0

Works cited

Fan, Jiaxin, Yafei Lyu, Qihuang Zhang, Xuran Wang, Mingyao Li, and Rui Xiao. 2022. “MuSiC2: Cell-Type Deconvolution for Multi-Condition Bulk RNA-Seq Data.” Briefings in Bioinformatics, October, bbac430. https://doi.org/10.1093/bib/bbac430.

Gong, Ting, and Joseph D. Szustakowski. 2013. “DeconRNASeq: A Statistical Framework for Deconvolution of Heterogeneous Tissue Samples Based on mRNA-Seq Data.” Bioinformatics 29 (8): 1083–5. https://doi.org/10.1093/bioinformatics/btt090.

Maden, Sean K., Sang Ho Kwon, Louise A. Huuki-Myers, Leonardo Collado-Torres, Stephanie C. Hicks, and Kristen R. Maynard. 2023. “Challenges and Opportunities to Computationally Deconvolve Heterogeneous Tissue with Varying Cell Sizes Using Single Cell RNA-Sequencing Datasets.” arXiv. https://doi.org/10.48550/arXiv.2305.06501.

Racle, Julien, and David Gfeller. 2020. “EPIC: A Tool to Estimate the Proportions of Different Cell Types from Bulk Gene Expression Data.” Edited by Sebastian Boegel, Methods in Molecular Biology,, 233–48. https://doi.org/10.1007/978-1-0716-0327-7_17.

Wang, Xuran, Jihwan Park, Katalin Susztak, Nancy R. Zhang, and Mingyao Li. 2019. “Bulk Tissue Cell Type Deconvolution with Multi-Subject Single-Cell Expression Reference.” Nature Communications 10 (1): 380. https://doi.org/10.1038/s41467-018-08023-x.