INLAomics is a hierarchical Bayesian model for analysing multiomic Spatial data using Integrated Nested Laplace Approximations (INLA). Biorxiv preprint INLAomics for Scalable and Interpretable Spatial Multiomic Data Integration.
All analysis in the manuscript is carried out in R V. 4.3.1 with packages R-INLA V. 23.12.17 and R-stan V. 2.26.23 (Stan V. 2.26.1). For instructions on installation we refer to R-INLA and mc-stan.
All models are implemented through the R-package INLA using the inla.rgeneric.define() method. The relevant scripts are under ./INLA/
In ./tutorial there is a script which gives some details of the implemented INLA methods and expected data formats.
The data generated in [1] is considered where files can be accessed at GSE198353. We have added cell annotations to two replicates of spleen tissue sections found in ./data.
The necessary files are
.
├── GSE198353_spleen_rep_1.csv
├── GSE198353_spleen_rep_1_filtered_feature_bc_matrix.h5
├── GSE198353_spleen_rep_2.csv
├── GSE198353_spleen_rep_2_filtered_feature_bc_matrix.h5
├── GSE198353_spleen_replicate_1_spatial.tar.gz
├── GSE198353_spleen_replicate_2_spatial.tar.gz
├── spatial
│ ├── qc_aligned_fiducials_image.jpg
│ ├── qc_detected_tissue_image.jpg
│ ├── scalefactors_json.json
│ ├── tissue_hires_image.png
│ ├── tissue_lowres_image.png
│ └── tissue_positions_list.csv
├── spatial2
│ ├── qc_aligned_fiducials_image.jpg
│ ├── qc_detected_tissue_image.jpg
│ ├── scalefactors_json.json
│ ├── tissue_hires_image.png
│ ├── tissue_lowres_image.png
│ └── tissue_positions_list.csv
...1_spatial.tar.gz and ...2_spatial.tar.gz are our own annotations.
Figure d above illustrates how INLAomoics can be utilized to construct genes-to-protein networks based on specific parameter estimates that encodes assay-to-assay effects.
Code to recreate CD3 rows are found in ./scripts/SPOTS/ProtVsGenes.R with runtime on Apple M2 approximately 6h. Instructions for recreating any of the other rows of Figure d is outlined in the script file.
To recreate the results of figure e & f, see ./scripts/SPOTS/SpleenPred.R
The necessary files are
.
├── GSE198353_mmtv_pymt.csv
├── GSE198353_mmtv_pymt_ADT.csv.gz
├── GSE198353_mmtv_pymt_GEX_filtered_feature_bc_matrix.h5
├── GSE198353_mmtv_pymt_spatial.tar.gz
└── spatial
├── aligned_fiducials.jpg
├── detected_tissue_image.jpg
├── scalefactors_json.json
├── tissue_hires_image.png
├── tissue_lowres_image.png
└── tissue_positions_list.csv
GSE198353_mmtv_pymt.csv are manual annotations found in ./data. Example code can be found in ./scripts/SPOTS/BreastPrediction.R.
.
├── raw_feature_bc_matrix
│ ├── barcodes.tsv.gz
│ ├── features.tsv.gz
│ └── matrix.mtx.gz
└── spatial
├── aligned_fiducials.jpg
├── aligned_tissue_image.jpg
├── cytassist_image.tiff
├── detected_tissue_image.jpg
├── scalefactors_json.json
├── spatial_enrichment.csv
├── tissue_hires_image.png
├── tissue_lowres_image.png
└── tissue_positions.csv
Example code can be found in ./scripts/visium/tonsil.R
.
├── GSM6578059_mousecolon_RNA.tsv.gz
├── GSM6578061_mousekidney_RNA.tsv.gz
├── GSM6578062_humantonsil_RNA.tsv.gz
├── GSM6578064_humanthymus_RNA.tsv.gz
├── GSM6578065_humanskin_RNA.tsv.gz
├── GSM6578068_mousecolon_protein.tsv.gz
├── GSM6578070_mousekidney_protein.tsv.gz
├── GSM6578071_humantonsil_protein.tsv.gz
├── GSM6578073_humanthymus_protein.tsv.gz
└── GSM6578074_humanskin_protein.tsv.gz
Example code can be found in ./scripts/Highplex/highplex.R
The scripts for carrying out the simulation studies are found in ./scripts/simulation/. The script for comparison between INLA and MCMC is outlined InlaVsMcmc.R, where one Monte-Carlo round takes approximately 30 minutes. The script for comparison between INLAomics and univariate correlation is found in InlaVsCorr.R, where one Monte-Carlo round takes approximately 25 seconds.
[1] Ben-Chetrit, N., Niu, X., Swett, A. D., Sotelo, J., Jiao, M. S., Stewart, C. M., ... & Landau, D. A. (2023). Integration of whole transcriptome spatial profiling with protein markers. Nature biotechnology, 41(6), 788-793.