Protocol for integrative transcriptomic/metabolomic analysis: A case study on factors influencing the oyster-POMS disease Authors: Leo Duperret, Alejandro Valdivieso, Bruno Petton, Benjamin Morga, Nicole Faury, Julien de Lorgeril, Charlotte Corporeau, Lionel Degremont, Sylvain Henry, Andrei Turtoi, Juliette Pouzadoux, Océane Romatif, Gaelle Courtay, Cristián J Monaco, Eve Toulza, Arnaud Lagorce, Jeremie Vidal-Dupiol, Guillaume Mitta Published: 2026 ABSTRACT SummaryDeciphering the molecular mechanisms underlying complex biological interactions requires comprehensive approaches that capture these interactions across multiple regulatory and functional levels. Here, we present a protocol applied to the host-pathogen-environment interactions underlying Pacific oyster mortality syndrome (POMS) in Magallana gigas. We describe steps from individual transcriptomic and metabolomic workflows using WGCNA and FELLA packages, to their integration in a biologically meaningful framework. This protocol aims to move omics integration beyond statistical correlations toward functional enrichment with cross-validation at the system level.For complete details on the use and execution of this protocol, please refer to Duperret et al.1Subject areas: Bioinformatics, RNAseq, Metabolomics Before you begin Pacific Oyster Mortality Syndrome (POMS) is a polymicrobial and multifactorial disease affecting the oyster Magallana gigas, a marine species of significant economic interest. While the molecular mechanisms underlying the polymicrobial manifestation of this disease has already been deciphered,2 the effect of food availability on POMS was recently elucidated. For this, we used two full-sib oyster families with distinct genetic backgrounds selected on their high susceptibility to POMS. At 3 months-old, “recipient” oysters were divided into two batches and followed a strict diet regime: fed ad libitum (permissive condition) or starved (non-permissive condition) for 16 days prior the infection. Infection was done in mesocosm, to mimic natural disease transmission using infected oysters (“donors”) injected with OsHV-1 μVar virus placed in the same environment as naïve recipient oysters. At different hours post-cohabitation (0, 3, 6, 12, 24, 48 and 72 hpc) a pool of ten oysters per family was sampled for RNA and metabolites extraction. RNA-sequencing was performed to analyze differentially expressed genes, and targeted metabolomics (LC/MS and GC/MS) was conducted to quantify relative metabolite abundance. Targetted metabolomic analysis quantified 346 metabolites across four metabolite classes: General Metabolites (GM, n = 107 covering amino acid metabolism, urea cycle, neurotransmitter metabolism, redox/cofactor metabolism); Nucleotides (n = 20); Poly-Unsaturated Fatty Acids (PUFAs, n = 151 including prostaglandins, leukotrienes, and oxylipins); and the Tricarboxylic Acid cycle (TCA) intermediates (n = 13). The list of targeted metabolites can be retrieved in Table S1: List of metabolites identified for the targeted metabolomics analysis. The protocol described here was focused on the combined analysis of RNA-seq and metabolomics data obtained in naïve oysters, after 16 days of acclimation just before POMS infection. The effect of food availability on the infection kineticis published in the original research paper.1 This protocol provides a comprehensive, end-to-end workflow, spanning from (i) single omics data analyses to (ii) the integration of both. Rather than relying solely on the mathematical correlations between omics, the pipeline combines functional enrichment with cross-validation to produce biologically meaningful, robust and reproducible results. This approach yields a systemic understanding of the underlying molecular mechanisms and biological interactions underlying complex biological processes. Innovation Previous studies have integrated multi-omics data by identifying statistical associations between large omics datasets (e.g., RNA-seq, metabolomics, proteomics), relying primarily on correlation-based approaches such as Pearson or Spearman coefficients. We propose here a strategy aimed at cross-validating outcomes from two independent omics analyses (RNA-seq and metabolomics), thereby integrating datasets with greater biological insight. This approach is based on (i) the identification of co-expressed genes and metabolites significantly associated to the phenotype of interest using the WGCNA pipeline; (ii) functional enrichment analyses of the identified features from each -omics, (iii) identification of the genes involved in the identified metabolic pathways, and (iv) the analysis of their respective biological role within these pathways associate with the phenotype. This rational enable to integrate cross-validated results from two omics layers. Institutional permissions Ethical approval was not required for this study in accordance with local legislation and institutional requirements, as there is no regulatory framework in Europe governing the use of bivalve molluscs in scientific research (cf. European Directive 2010/63/EU on the protection of animals used for scientific purposes). Nevertheless, the 3R principle (Replacement, Reduction, Refinement) was applied throughout the study,1 notably by minimizing the number of animals used, and were anesthetized prior to manipulation in order to minimize potential stress. Pre-processing of data for RNA-sequencing analysis Timing: 1–3 days 1.Create a free account in Galaxy France (https://usegalaxy.fr/). Create a free account in Galaxy France (https://usegalaxy.fr/). 2.Upload raw RNA seq data files in a new Galaxy history. Upload raw RNA seq data files in a new Galaxy history. 3.Check read quality with FastQC (v.0.7.1).a.Raw read data from your current history: raw data.fastq.gz.b.Contaminant list: default parameters.c.Adapter list: default parameters.d.Submodule and Limit specifying file: default parameters.e.Disable grouping of bases for reads > 50 bp: No.f.Length of Kmer: 7. Check read quality with FastQC (v.0.7.1).a.Raw read data from your current history: raw data.fastq.gz.b.Contaminant list: default parameters.c.Adapter list: default parameters.d.Submodule and Limit specifying file: default parameters.e.Disable grouping of bases for reads > 50 bp: No.f.Length of Kmer: 7. a.Raw read data from your current history: raw data.fastq.gz. Raw read data from your current history: raw data.fastq.gz. b.Contaminant list: default parameters. Contaminant list: default parameters. c.Adapter list: default parameters. Adapter list: default parameters. d.Submodule and Limit specifying file: default parameters. Submodule and Limit specifying file: default parameters. e.Disable grouping of bases for reads > 50 bp: No. Disable grouping of bases for reads > 50 bp: No. f.Length of Kmer: 7. Length of Kmer: 7. 4.Trim low quality/short reads and remove adapters with Trim Galore! (v.0.6.3).a.According to sequencing design select paired- or singled-end.b.Reads in FASTQ format: raw data.fastq.gz.c.Adapter sequence to be trimmed: User defined adapter sequence.d.Trim Galore! advanced settings: Full parameters list.i.Trim low-quality ends from reads in addition to adapter removal: 30.ii.Overlap with adapter sequence required to trim a sequence: 1.iii.Maximum allowed error rate: 0.1.iv.Discard reads that became shorter than length N: 35.v.Generate a report file: Yes.vi.According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads.e.RBBS specific settings: default parameters (no RBBS).f.Trimming settings: default parameters. Trim low quality/short reads and remove adapters with Trim Galore! (v.0.6.3).a.According to sequencing design select paired- or singled-end.b.Reads in FASTQ format: raw data.fastq.gz.c.Adapter sequence to be trimmed: User defined adapter sequence.d.Trim Galore! advanced settings: Full parameters list.i.Trim low-quality ends from reads in addition to adapter removal: 30.ii.Overlap with adapter sequence required to trim a sequence: 1.iii.Maximum allowed error rate: 0.1.iv.Discard reads that became shorter than length N: 35.v.Generate a report file: Yes.vi.According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads.e.RBBS specific settings: default parameters (no RBBS).f.Trimming settings: default parameters. a.According to sequencing design select paired- or singled-end. According to sequencing design select paired- or singled-end. b.Reads in FASTQ format: raw data.fastq.gz. Reads in FASTQ format: raw data.fastq.gz. c.Adapter sequence to be trimmed: User defined adapter sequence. Adapter sequence to be trimmed: User defined adapter sequence. d.Trim Galore! advanced settings: Full parameters list.i.Trim low-quality ends from reads in addition to adapter removal: 30.ii.Overlap with adapter sequence required to trim a sequence: 1.iii.Maximum allowed error rate: 0.1.iv.Discard reads that became shorter than length N: 35.v.Generate a report file: Yes.vi.According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads. Trim Galore! advanced settings: Full parameters list.i.Trim low-quality ends from reads in addition to adapter removal: 30.ii.Overlap with adapter sequence required to trim a sequence: 1.iii.Maximum allowed error rate: 0.1.iv.Discard reads that became shorter than length N: 35.v.Generate a report file: Yes.vi.According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads. i.Trim low-quality ends from reads in addition to adapter removal: 30. Trim low-quality ends from reads in addition to adapter removal: 30. ii.Overlap with adapter sequence required to trim a sequence: 1. Overlap with adapter sequence required to trim a sequence: 1. iii.Maximum allowed error rate: 0.1. Maximum allowed error rate: 0.1. iv.Discard reads that became shorter than length N: 35. Discard reads that became shorter than length N: 35. v.Generate a report file: Yes. Generate a report file: Yes. vi.According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads. According to library sequencing design specify if you would like to retain unpaired reads: Do not output unpaired reads or output unpaired reads. e.RBBS specific settings: default parameters (no RBBS). RBBS specific settings: default parameters (no RBBS). f.Trimming settings: default parameters. Trimming settings: default parameters. 5.Map reads to the reference genome with RNA STAR (v.2.6.0b-1).a.According to library sequencing design: Single-end or Paired-end reads.b.RNA-seq FASTQ/FASTA file: from Trim Galore!c.Custom or built-in reference genome: Use reference genome from history and create temporary index.i.Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1).ii.Length of the SA pre-indexing string: 14.iii.Build index with or without known splice junctions annotation: build index without gene-model.d.Use 2-pass mapping for sensitive novel splice junction discovery: No.e.Per gene/transcript output: No per gene or transcript output.f.Report chimeric alignments: Don’t report chimeric alignments.g.BAM output format specification:i.Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment).ii.HI tag values should be: one-based.iii.MAPQ value for unique mappers: 60.h.Output filter criteria:i.Exclude the following records from the BAM output: unselect all.ii.Would you like to set additional output filters: No.i.Algorithmic settings_Configure seed, alignment and limits options: Use defaults. Map reads to the reference genome with RNA STAR (v.2.6.0b-1).a.According to library sequencing design: Single-end or Paired-end reads.b.RNA-seq FASTQ/FASTA file: from Trim Galore!c.Custom or built-in reference genome: Use reference genome from history and create temporary index.i.Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1).ii.Length of the SA pre-indexing string: 14.iii.Build index with or without known splice junctions annotation: build index without gene-model.d.Use 2-pass mapping for sensitive novel splice junction discovery: No.e.Per gene/transcript output: No per gene or transcript output.f.Report chimeric alignments: Don’t report chimeric alignments.g.BAM output format specification:i.Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment).ii.HI tag values should be: one-based.iii.MAPQ value for unique mappers: 60.h.Output filter criteria:i.Exclude the following records from the BAM output: unselect all.ii.Would you like to set additional output filters: No.i.Algorithmic settings_Configure seed, alignment and limits options: Use defaults. a.According to library sequencing design: Single-end or Paired-end reads. According to library sequencing design: Single-end or Paired-end reads. b.RNA-seq FASTQ/FASTA file: from Trim Galore! RNA-seq FASTQ/FASTA file: from Trim Galore! c.Custom or built-in reference genome: Use reference genome from history and create temporary index.i.Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1).ii.Length of the SA pre-indexing string: 14.iii.Build index with or without known splice junctions annotation: build index without gene-model. Custom or built-in reference genome: Use reference genome from history and create temporary index.i.Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1).ii.Length of the SA pre-indexing string: 14.iii.Build index with or without known splice junctions annotation: build index without gene-model. i.Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1). Select a reference genome: Crassotrea_gigas_Roslin_v1_genomic.fa (cgigas_uk_roslin_v1, Assembly: GCA902806645.1). ii.Length of the SA pre-indexing string: 14. Length of the SA pre-indexing string: 14. iii.Build index with or without known splice junctions annotation: build index without gene-model. Build index with or without known splice junctions annotation: build index without gene-model. d.Use 2-pass mapping for sensitive novel splice junction discovery: No. Use 2-pass mapping for sensitive novel splice junction discovery: No. e.Per gene/transcript output: No per gene or transcript output. Per gene/transcript output: No per gene or transcript output. f.Report chimeric alignments: Don’t report chimeric alignments. Report chimeric alignments: Don’t report chimeric alignments. g.BAM output format specification:i.Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment).ii.HI tag values should be: one-based.iii.MAPQ value for unique mappers: 60. BAM output format specification:i.Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment).ii.HI tag values should be: one-based.iii.MAPQ value for unique mappers: 60. i.Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment). Read alignment tags to include in the BAM output: select NH (number of reported alignments/hits for the read), HI (query hit index), AS (local alignment score), nM (number of mismatches per (paired) alignment), ch (used to indicate chimeric alignment). ii.HI tag values should be: one-based. HI tag values should be: one-based. iii.MAPQ value for unique mappers: 60. MAPQ value for unique mappers: 60. h.Output filter criteria:i.Exclude the following records from the BAM output: unselect all.ii.Would you like to set additional output filters: No. Output filter criteria:i.Exclude the following records from the BAM output: unselect all.ii.Would you like to set additional output filters: No. i.Exclude the following records from the BAM output: unselect all. Exclude the following records from the BAM output: unselect all. ii.Would you like to set additional output filters: No. Would you like to set additional output filters: No. i.Algorithmic settings_Configure seed, alignment and limits options: Use defaults. Algorithmic settings_Configure seed, alignment and limits options: Use defaults. 6.Count the number of mapped reads to each gene with htseq-count (v.0.9.1).a.Aligned SAM/BAM file: from RNA STAR output.b.GFF file: genome_ref_Cgigas.gff3.c.Mode: Union.d.According to library preparation strategy Stranded: Reverse (stranded library).e.Minimum alignment quality: 10.f.Feature type: gene.g.ID attribute: ID. Count the number of mapped reads to each gene with htseq-count (v.0.9.1).a.Aligned SAM/BAM file: from RNA STAR output.b.GFF file: genome_ref_Cgigas.gff3.c.Mode: Union.d.According to library preparation strategy Stranded: Reverse (stranded library).e.Minimum alignment quality: 10.f.Feature type: gene.g.ID attribute: ID. a.Aligned SAM/BAM file: from RNA STAR output. Aligned SAM/BAM file: from RNA STAR output. b.GFF file: genome_ref_Cgigas.gff3. GFF file: genome_ref_Cgigas.gff3. c.Mode: Union. Mode: Union. d.According to library preparation strategy Stranded: Reverse (stranded library). According to library preparation strategy Stranded: Reverse (stranded library). e.Minimum alignment quality: 10. Minimum alignment quality: 10. f.Feature type: gene. Feature type: gene. g.ID attribute: ID. ID attribute: ID. 7.Set advanced options: Default settings. Set advanced options: Default settings. Pre-processing of data for targeted metabolomic analysis Timing: 1 day 8.Normalize all the metabolites’ MS peak area counts by their respective internal standard. Normalize all the metabolites’ MS peak area counts by their respective internal standard. Note: To do so, in the excel file with your raw data, divide each metabolite’s area count by the area count of the internal standard. In case you have one excel file per targeted metabolomic analysis, combine all individual files to create one dataset containing every normalized metabolite per sample. 9.Pre-process each sample in MetaboAnalystR (v.4.0.0) in R (v4.3.2) for normalization and data transformation. Pre-process each sample in MetaboAnalystR (v.4.0.0) in R (v4.3.2) for normalization and data transformation. Note: You could also use the online website (https://www.metaboanalyst.ca/). >install.packages(“pacman”) >install.packages(“devtools”) >library(pacman) >library(devtools) >pacman::p_load(c(“impute”, “pcaMethods”, “globaltest”, “GlobalAncova”, “Rgraphviz”, “preprocessCore”, “genefilter”, “sva”, “limma”, “KEGGgraph”, “siggenes”, “BiocParallel”, “MSnbase”, “multtest”,“RBGL”, “edgeR”, “fgsea”, “httr”, “qs”)) >devtools::install_github(“xia-lab/MetaboAnalystR”, build = TRUE, build_vignettes = TRUE, build_manual =T) >library(MetaboAnalystR) >mSet<-InitDataObjects(“pktable”, “stat”, FALSE) >mSet<-Read.TextData(mSet, “Replacing_with_your_file_path”, “colu”, “disc”); >mSet<-SanityCheckData(mSet) >mSet<-ReplaceMin(mSet) >mSet<-FilterVariable(mSet, “F”, 25, “iqr”, 10, “mean”, 0) >mSet<-PreparePrenormData(mSet) >mSet<-Normalization(mSet, “QuantileNorm”, “LogNorm”, “NULL”, ratio=FALSE, ratioNum=20) >data.norm <- mSet[[“dataSet”]][[ “norm”]] %>% t() CRITICAL: If you can not find a particular library on R, you could try running this code line: >install.packages(“name of the package”). If it is still not working because it can not be found, please refer to the documentation on GitHub of the specific tool. Note: Use the “pktable” command since the data represents Peak intensities, “colu” since the samples are in column (unpaired), “F” and 25 to filter out metabolites with less than 25% of variance, “iqr” and 10 to remove the 10% least variable metabolites based on the interquartile range, “QuantileNorm” and “LogNorm” for a Quantile normalization followed by a Logarithmic transformation. The percentage of metabolite to be filtered out follows the empirical rule as follow: 1) Less than 250 variables: 5% will be filtered; 2) Between 250–500 variables: 10% will be filtered; 3) Between 500–1000 variables: 25% will be filtered; and 4) Over 1000 variables: 40% will be filtered; based on MetaboAnalyst tutorial. The ReplaceMin function replaces zero and missing values with half of the smallest positive value in the dataset, enabling subsequent log-normalization. At this point, the transcriptomic and metabolomic datasets are ready for further integrated analysis. Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited dataTranscriptomics raw dataDuperret et al.1Bioproject DECICOMP: PRJEB86647Metabolomics raw dataDuperret et al.1MTBLS 13313Software and algorithmsLaptopDELL, Precision 3561, Intel Core i7, 16 Go RAMGalaxyGoecks et al.3https://usegalaxy.fr/FastQC version 0.7.1Andrews.4Trim Galore! version 0.6.3Krueger.5RNA STAR version 2.6.0b-1Dobin et al.6htseq-count version 0.9.1Anders et al.7MetaboAnalyst version 4.0.0Chong et al.8https://www.metaboanalyst.ca/R version 4.3.2R Core Team.9https://www.r-project.org/DESeq2 version 1.42.0Love et al.10Weighted Gene Correlation Analysis (WGCNA) version 1.72–5Langfelder et al.11https://cran.r-project.org/web/packages/WGCNA/index.htmlKEGGKanehisa et al.12https://www.genome.jp/kegg/kegg2.htmlFELLA version 1.22.0Picart-Armada et al.13https://rdrr.io/bioc/FELLA/man/FELLA.htmlDAVID_BioinformaticSherman et al.,14 Huang et al.15https://davidbioinformatics.nih.gov/NCBIN/Ahttps://www.ncbi.nlm.nih.gov/datasets/gene/Experimental models: Organisms/strainsOyster Magallana gigasIFREMER (France)N/A Step-by-step method details Individual omics analysis: Computational analysis of the transcriptomic data Timing: 1–3 days The integrative multi-omics analysis starts with an omics-by-omics bio-analysis treatment. Here we describe how to analyze the transcriptomics data to identify genes correlated to a specific trait/phenotype of interest based on the use of DESeq2 (v. 1.42.0) and Weighted Gene Correlation Network Analysis (WGCNA, v. 1.72–5) packages. These results will be the input for the multi-omics integration. 1.Select the raw matrix/dataframe counts from the samples. Select the raw matrix/dataframe counts from the samples. Note: In our case, we selected only the samples corresponded to the basal condition (T0) (NT = 12 samples; 2 families x 3 biological replicates x 2 diet conditions). 2.Perform a quality control on the data and remove genes with no variance across samples;and genes/samples with missing values. Perform a quality control on the data and remove genes with no variance across samples;and genes/samples with missing values. >install.packages(“WGCNA”, dependencies = TRUE) >library(WGCNA) >countData <- read your document containing your raw transcriptomics counts >gsg <- goodSamplesGenes(t(countData)) >table(gsg$goodGenes) >table(gsg$goodSamples) >countData <- countData[gsg$goodGenes==TRUE,] >coldata <- read your document containing the metadata >all(rownames(coldata) %in% colnames(countData)) >all(rownames(coldata) == colnames(countData)) 3.Ensure that the metadata samples associated to the countData have the exact names and number size. Ensure that the metadata samples associated to the countData have the exact names and number size. 4.Filter low expressed genes and apply Variance Stabilizing Transformation (VST). Filter low expressed genes and apply Variance Stabilizing Transformation (VST). >if (!requireNamespace(“BiocManager”, quietly = TRUE)){install.packages(“BiocManager”)} >BiocManager::install(“DESeq2″) >library(DESeq2) >dds <- DESeqDataSetFromMatrix(countData = countData, colData = coldata, design = ∼1) >dds <- dds[rowSums(counts(dds) >= 15) >= 9, ] >dds_norm <- vst(dds) Note: The design is set to ∼1 because this step serves to prepare the VST transformation. We do not want any group comparison at this stage. The WGCNA FAQ recommends to remove features whose counts are consistently low to reduce noise. We decided to exclud genes with counts < 15 in more than 75% of samples (hence 9; 0.75∗12=9). We applied library normalization and then the VST transformation. 5.Chose the soft power threshold (see troubleshooting, problem 2). Chose the soft power threshold (see troubleshooting, problem 2). >dds_norm <- assay(dds_norm) %>% t() >power <- c(c(1:10), seq(from =1, to =30, by =1 )) >sft <- pickSoftThreshold(dds_norm, powerVector = power, networkType = “signed”, RsquaredCut = 0.95, verbose = 5) >sft.data <- sft$fitIndices >a1 <- ggplot(sft.data, aes(Power, SFT.R.sq, label=Power)) + geom_point() + geom_text(nudge_y =0.1) + geom_hline(yintercept = 0.8, color=“red”)+ labs(x=“Power”, y= “scale free topology model filt, signed Rˆ2″)+ theme_classic() >a2 <- ggplot(sft.data, aes(Power, mean.k., label=Power)) + geom_point() +geom_text(nudge_y =0.1) + geom_hline(yintercept = 0.8, color=“red”) +labs(x=“Power”, y= “Mean Connectivity”)+ theme_classic() >grid.arrange(a1, a2, nrow=2) CRITICAL: To construct a scale-free topology network, a soft power threshold (sft) needs to have a R2 above 0.8 (above the red line, see Figure 1). 6.Compute the scale-free topology unsupervised signed network. Compute the scale-free topology unsupervised signed network. >dds_norm[] <- sapply(dds_norm, as.numeric) >soft_power <- # select your soft power >temp_cor <- cor >cor <- WGCNA::cor >bwnet <- blockwiseModules(dds_norm, maxBlockSize = # number max of genes contained in one module, networkType = “signed”, TOMType = “signed”, power = soft_power, mergeCutHeight = # modules whose eigengene dissimilarity is below this threshold will be merged, minModuleSize= # number min of genes contained in one module, numericLabels = FALSE, randomSeed=1234, nThreads = 4, verbose =3) >cor <- temp_cor 7.Identify the modules that are significantly correlated to your trait of interest -variable you are interested in (see Figure 2 and troubleshooting, problem 4). Identify the modules that are significantly correlated to your trait of interest -variable you are interested in (see Figure 2 and troubleshooting, problem 4). >module_eigengenes <- bwnet$Mes >coldata <- coldata %>% mutate(#name of your trait = case_when(#trait == “XX” ∼ as.character(1), #trait ==“YY” ∼ as.character(2), TRUE ∼ NA_character_)) >traits <- coldata %>% dplyr::select(#name of your trait) >nSamples <- nrow(dds_norm) >nGenes <- ncol(dds_norm) >module.trait.cor <- cor(module_eigengenes, traits, use = ‘p') >module.trait.cor.pvalues <- corPvalueStudent(module.trait.cor, nSamples) >heatmap.data <- merge(module_eigengenes, traits, by ='row.names') >heatmap.data <- heatmap.data %>% column_to_rownames(var='Row.names') >Corrrelations_color_permissive <- CorLevelPlot(heatmap.data, x = names(heatmap.data)[# columns containing your trait of interest(s)], y = names(heatmap.data)[# columns containing modules], col=c(“blue1″, “skyblue”, “white”, “pink”, “red”)) >Corrrelations_color_permissive Note: The grey module contains all genes that are not assigned to any co-expression module (e.g., genes showing low correlation with other genes in the dataset or belonging to clusters that are too small). It should be discarded for downstream analyses. cor() is used to calculate the Pearson correlation coefficient between each gene and the corresponding module eigengene. 8.Calculate the module membership and associated p-value of each gene relative to its assigned module eigengenes. Calculate the module membership and associated p-value of each gene relative to its assigned module eigengenes. >module.membership.measure <- cor(module_eigengenes, dds_norm, use='p') >module.membership.measure.pvalues <- corPvalueStudent(module.membership.measure, nSamples) >module.membership.measure.pvalues <- as.data.frame(t(module.membership.measure.pvalues)) >gene.signf.corr <- cor(dds_norm, traits$permissive, use='p') >gene.signf.corr.pvals <- corPvalueStudent(gene.signf.corr, nSamples) >geneModuleMembership <- as.data.frame(cor(dds_norm, module_eigengenes, use='p')) 9.Retrieve the list of genes belonging to significantly correlated modules associate with the phenotype, together with their module membership values. Retrieve the list of genes belonging to significantly correlated modules associate with the phenotype, together with their module membership values. >module.gene.mapping <- data.frame(cluster = as.character(bwnet$colors), Gene = names(bwnet$colors), stringsAsFactors = FALSE) >colour of the module <- module.gene.mapping %>% filter(cluster == “colour”) >kME.module <- geneModuleMembership %>% dplyr::select(.,ME#colour) %>% filter(., rownames(.) %in% #colour$Gene) >colnames(kME.module)[1] <- “kME” Note: Here, the “#colour” stands for the name of the module. The name is a colour name (e.g. MEblack, MEcyan). Selection of the soft power threshold to achieve a scale-free topology for WGCNA analyses A scale-free topology is characterized by a network structure in which most features (genes/metabolites) have few connections, whereas a small proportion of highly connected features act as hubs. Here, when applied to the transcriptomics data of,1 the red horizontal line indicates the threshold of 0.8 for the signed scale-free topology fit index, and a soft-thresholding power of 27 was selected for subsequent network construction. Modules correlated to your trait of interest, containing co-expressed features (genes/metabolites), identified by WGCNA Heatmap showing both the Pearson correlation coefficient from the module to the trait of interest (top), and the corresponding P-value (bottom). Colors represents the strength and the direction of the correlation, with red and blue indicating positive and negative correlations, respectively. Here, when applied to the transcriptomics data of,1 the modules brown, blue and turquoise shows significant correlation to the traits, thus selected for further analyses. The figure is the Supplementary Figure 3-A from Duperret et al.1 Individual omics analysis: Computational analysis of the metabolomic data Timing: 1–3 days Here we present a two-step method to individually analyze the metabolomics data: (i) extraction of the list of metabolites correlated to a specific trait of interest with Weighted Gene (metabolite) Correlation Network Analysis (WGCNA) package; (ii) the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis with Fella package (v. 1.22.0). These results will be the input for the multi-omics integration. Note: Follow the same protocol used for the transcriptomics data (Step 1–9) except for: (i) skip the Step 4 since metabolomics data have already been normalized (cf Pre-processing of data for targeted metabolomics analysis) ; (ii) you may need to reduce the minModuleSize in the Step 6 to match with your sample size. 10.Convert your metabolite of interest into universal KEGG compound identifier. Convert your metabolite of interest into universal KEGG compound identifier. Note: Type your metabolite in the browser of https://www.genome.jp/kegg/kegg2.html and select the corresponding KEGG COMPOUND. 11.Create a KEGG metabolic sub-network specific of your organism. Create a KEGG metabolic sub-network specific of your organism. > if (!requireNamespace(“BiocManager”, quietly = TRUE)) { install.packages(“BiocManager”) } >BiocManager::install(“FELLA”) >library(“FELLA”) >graph <- buildGraphFromKEGGREST(organism = “#your organism code”) >tmpdir <- paste0(tempdir(),“/my_database”) >unlink(tmpdir, recursive = TRUE) >buildDataFromGraph(keggdata.graph = graph, databaseDir = tmpdir, internalDir = FALSE, matrices = c(“diffusion”), normality = c(“diffusion”), niter = 1e3) >fella.data <- loadKEGGdata(databaseDir = tmpdir, internalDir = FALSE, loadMatrix = c(“diffusion”)) >cat(getInfo(fella.data)) 12.Compute the metabolic enrichment analysis with your metabolites of interest extracted from WGCNA analysis. Compute the metabolic enrichment analysis with your metabolites of interest extracted from WGCNA analysis. >analysis <- defineCompounds(compounds = #your compound dataframe, data = fella.data) >getInput(analysis) >getExcluded(analysis) >analysis <- runDiffusion(object = analysis, data = fella.data, approx = “normality”) > exportResults(format = “csv”, file = “Your_file.csv”, method = “diffusion”, object = analysis, data = fella.data, nlimit = 1000) 13.Extract the metabolic pathway differentially represented based on the metabolites of interest. Extract the metabolic pathway differentially represented based on the metabolites of interest. >metabo_results <- read.csv(“Your_file.csv”) >pathway_results <- metabo_results %>% filter(Entry.type==“pathway”) Integration of multi-omics analysis Timing: 1 day The final step of this analysis is to integrate both omic (transcriptomics and metabolomics) in a biologically meaningful way. The rational developed in,1 was to select genes that were both (i) co-expressed and significantly associated with the trait of interest (WGCNA analysis; Step 1–9); and (ii) involved in the same metabolite-associated pathways identified through functional enrichment of the metabolite correlation network (WGCNA and FELLA analyses; Step 10–14). 14.Convert gene identifiers from the extracted gene list into the KEGG gene ID format.Note: To do so, create a conversion file containing the correspondence between the identifier used in your organism’s genome annotation (hereafter called Genes_Organism) with the KEGG gene ID format (hereafter called Genes_KEGG), in case is needed.CRITICAL: Do this step only if the gene identifier from your organism is not a KEGG gene ID formata.Download all the KEGG information of your organism (pathways, compounds, genes)>if (!requireNamespace(“BiocManager”, quietly = TRUE)) {install.packages(“BiocManager”)}>BiocManager::install(“EnrichmentBrowser”)>library(EnrichmentBrowser)>crg_pathways <- downloadPathways(“#your organism code”)>crg_pathways_individual <- getGenesets(org = “#your organism KEGG code ”, db = “kegg”, cache = TRUE, return.type=“list”)b.Extract the genes involved in the KEGG map of your organism.>pathway_dfs <- list()>for (i in seq_along(crg_pathways_individual)) { kegg_pathway <- names(crg_pathways_individual)[i] pathway_genes <- crg_pathways_individual[[i]] pathway_df <- data.frame(Gene = pathway_genes, kegg_pathway = kegg_pathway) pathway_df$code <- substring(kegg_pathway, 1, 8) pathway_dfs[[i]] <- pathway_df}>All_data_kegg_pathways_genes <- bind_rows(pathway_dfs) %>% mutate(kegg_pathway = substr(kegg_pathway, 9, nchar(kegg_pathway)), kegg_pathway = gsub(“_”, “”, kegg_pathway))c.Convert the Genes_KEGG obtained in (b) to match with Genes_Organism by using DAVID_Bioinformatic (https://davidbioinformatics.nih.gov/).d.For the Genes_Organism that did not have a match with the Genes_KEGG obtained in (b), find a match using NCBI (https://www.ncbi.nlm.nih.gov/datasets/gene/).e.For the remaining gene identifiers, do a BlastN from the Genes_KEGG to Genes_Organism.f.For the genes that had either an alignment to multiple Genes_Organism, or for the ones that had a short length alignment, do a BlastP and select the best match.g.Finally, the Genes_KEGG that had a different match between the BlastN and the BlastP and for the ones that had an ID% < 90%, discard the gene. Convert gene identifiers from the extracted gene list into the KEGG gene ID format.Note: To do so, create a conversion file containing the correspondence between the identifier used in your organism’s genome annotation (hereafter called Genes_Organism) with the KEGG gene ID format (hereafter called Genes_KEGG), in case is needed.CRITICAL: Do this step only if the gene identifier from your organism is not a KEGG gene ID formata.Download all the KEGG information of your organism (pathways, compounds, genes)>if (!requireNamespace(“BiocManager”, quietly = TRUE)) {install.packages(“BiocManager”)}>BiocManager::install(“EnrichmentBrowser”)>library(EnrichmentBrowser)>crg_pathways <- downloadPathways(“#your organism code”)>crg_pathways_individual <- getGenesets(org = “#your organism KEGG code ”, db = “kegg”, cache = TRUE, return.type=“list”)b.Extract the genes involved in the KEGG map of your organism.>pathway_dfs <- list()>for (i in seq_along(crg_pathways_individual)) { kegg_pathway <- names(crg_pathways_individual)[i] pathway_genes <- crg_pathways_individual[[i]] pathway_df <- data.frame(Gene = pathway_genes, kegg_pathway = kegg_pathway) pathway_df$code <- substring(kegg_pathway, 1, 8) pathway_dfs[[i]] <- pathway_df}>All_data_kegg_pathways_genes <- bind_rows(pathway_dfs) %>% mutate(kegg_pathway = substr(kegg_pathway, 9, nchar(kegg_pathway)), kegg_pathway = gsub(“_”, “”, kegg_pathway))c.Convert the Genes_KEGG obtained in (b) to match with Genes_Organism by using DAVID_Bioinformatic (https://davidbioinformatics.nih.gov/).d.For the Genes_Organism that did not have a match with the Genes_KEGG obtained in (b), find a match using NCBI (https://www.ncbi.nlm.nih.gov/datasets/gene/).e.For the remaining gene identifiers, do a BlastN from the Genes_KEGG to Genes_Organism.f.For the genes that had either an alignment to multiple Genes_Organism, or for the ones that had a short length alignment, do a BlastP and select the best match.g.Finally, the Genes_KEGG that had a different match between the BlastN and the BlastP and for the ones that had an ID% < 90%, discard the gene. Note: To do so, create a conversion file containing the correspondence between the identifier used in your organism’s genome annotation (hereafter called Genes_Organism) with the KEGG gene ID format (hereafter called Genes_KEGG), in case is needed. CRITICAL: Do this step only if the gene identifier from your organism is not a KEGG gene ID format a.Download all the KEGG information of your organism (pathways, compounds, genes)>if (!requireNamespace(“BiocManager”, quietly = TRUE)) {install.packages(“BiocManager”)}>BiocManager::install(“EnrichmentBrowser”)>library(EnrichmentBrowser)>crg_pathways <- downloadPathways(“#your organism code”)>crg_pathways_individual <- getGenesets(org = “#your organism KEGG code ”, db = “kegg”, cache = TRUE, return.type=“list”) Download all the KEGG information of your organism (pathways, compounds, genes)>if (!requireNamespace(“BiocManager”, quietly = TRUE)) {install.packages(“BiocManager”)}>BiocManager::install(“EnrichmentBrowser”)>library(EnrichmentBrowser)>crg_pathways <- downloadPathways(“#your organism code”)>crg_pathways_individual <- getGenesets(org = “#your organism KEGG code ”, db = “kegg”, cache = TRUE, return.type=“list”) >if (!requireNamespace(“BiocManager”, quietly = TRUE)) {install.packages(“BiocManager”)} >BiocManager::install(“EnrichmentBrowser”) >library(EnrichmentBrowser) >crg_pathways <- downloadPathways(“#your organism code”) >crg_pathways_individual <- getGenesets(org = “#your organism KEGG code ”, db = “kegg”, cache = TRUE, return.type=“list”) b.Extract the genes involved in the KEGG map of your organism.>pathway_dfs <- list()>for (i in seq_along(crg_pathways_individual)) { kegg_pathway <- names(crg_pathways_individual)[i] pathway_genes <- crg_pathways_individual[[i]] pathway_df <- data.frame(Gene = pathway_genes, kegg_pathway = kegg_pathway) pathway_df$code <- substring(kegg_pathway, 1, 8) pathway_dfs[[i]] <- pathway_df}>All_data_kegg_pathways_genes <- bind_rows(pathway_dfs) %>% mutate(kegg_pathway = substr(kegg_pathway, 9, nchar(kegg_pathway)), kegg_pathway = gsub(“_”, “”, kegg_pathway)) Extract the genes involved in the KEGG map of your organism.>pathway_dfs <- list()>for (i in seq_along(crg_pathways_individual)) { kegg_pathway <- names(crg_pathways_individual)[i] pathway_genes <- crg_pathways_individual[[i]] pathway_df <- data.frame(Gene = pathway_genes, kegg_pathway = kegg_pathway) pathway_df$code <- substring(kegg_pathway, 1, 8) pathway_dfs[[i]] <- pathway_df}>All_data_kegg_pathways_genes <- bind_rows(pathway_dfs) %>% mutate(kegg_pathway = substr(kegg_pathway, 9, nchar(kegg_pathway)), kegg_pathway = gsub(“_”, “”, kegg_pathway)) >pathway_dfs <- list() >for (i in seq_along(crg_pathways_individual)) { kegg_pathway <- names(crg_pathways_individual)[i] pathway_genes <- crg_pathways_individual[[i]] pathway_df <- data.frame(Gene = pathway_genes, kegg_pathway = kegg_pathway) pathway_df$code <- substring(kegg_pathway, 1, 8) pathway_dfs[[i]] <- pathway_df} >All_data_kegg_pathways_genes <- bind_rows(pathway_dfs) %>% mutate(kegg_pathway = substr(kegg_pathway, 9, nchar(kegg_pathway)), kegg_pathway = gsub(“_”, “”, kegg_pathway)) c.Convert the Genes_KEGG obtained in (b) to match with Genes_Organism by using DAVID_Bioinformatic (https://davidbioinformatics.nih.gov/). Convert the Genes_KEGG obtained in (b) to match with Genes_Organism by using DAVID_Bioinformatic (https://davidbioinformatics.nih.gov/). d.For the Genes_Organism that did not have a match with the Genes_KEGG obtained in (b), find a match using NCBI (https://www.ncbi.nlm.nih.gov/datasets/gene/). For the Genes_Organism that did not have a match with the Genes_KEGG obtained in (b), find a match using NCBI (https://www.ncbi.nlm.nih.gov/datasets/gene/). e.For the remaining gene identifiers, do a BlastN from the Genes_KEGG to Genes_Organism. For the remaining gene identifiers, do a BlastN from the Genes_KEGG to Genes_Organism. f.For the genes that had either an alignment to multiple Genes_Organism, or for the ones that had a short length alignment, do a BlastP and select the best match. For the genes that had either an alignment to multiple Genes_Organism, or for the ones that had a short length alignment, do a BlastP and select the best match. g.Finally, the Genes_KEGG that had a different match between the BlastN and the BlastP and for the ones that had an ID% < 90%, discard the gene. Finally, the Genes_KEGG that had a different match between the BlastN and the BlastP and for the ones that had an ID% < 90%, discard the gene. 15.Use this data file to convert the gene identifier from the gene of interest extracted from Step 9. Use this data file to convert the gene identifier from the gene of interest extracted from Step 9. 16.From the list of genes of interest extracted from Step 9, select only the ones that are involved in the metabolic pathway differentially enriched from Step 15 (see Figure 3). From the list of genes of interest extracted from Step 9, select only the ones that are involved in the metabolic pathway differentially enriched from Step 15 (see Figure 3). Representation of the transcriptomics/metabolomics integration The final heatmap (on the right) represents the list of genes being (i) differentially expressed by the trait of interest - here the Diet - identified with WGCNA (heatmap on the left) and (ii) involved in the differentially represented metabolic pathway identified with FELLA based on the metabolites identified with WGCNA (dot plot on the bottom). With this integration, 967 out of the 13238 genes correlated to the specific trait of interest were involved in the metabolic pathways differentially represented based on the trait. >Specific_pathway <- All_data_kegg_pathways_genes %>% filter(code %in% pathway_results$KEGG.id) %>% merge(# file containing the gene ID of interest in the KEGG format, by=“Gene”) %>% dplyr::select(-5) 17.Visualise in KEGG (https://www.kegg.jp/) the implication of the genes in every metabolic pathway. Visualise in KEGG (https://www.kegg.jp/) the implication of the genes in every metabolic pathway. Expected outcomes The integration of both transcriptomics and metabolomics data generated a more comprehensive understanding of the studied phenomenon. Instead of relying on a single biological layer, this approach provides a deeper insight into the biological system while cross-validating results obtained at the transcriptome and metabolome level. This integrative strategy is expected to identify genes involved in metabolite-associated pathways, thereby highlighting coordinated molecular responses consistent with the observed metabolic changes and physiological state of the animal. Limitations This multi-omics integration can be applied on every organism at the condition that their metabolic map have been characterized and made publicly available on the KEGG database. Since the datasets used include metabolomic data, this protocol can be useful for integration with other omics dataset (epigenetic, genetic, proteomic), enabling the identification of epigenetics/genes/proteins involved in the studied phenomenon. Any omics dataset could be integrated, as long as a biologically meaningful link can be found. In the case of this protocol (RNA-seq and metabolomics), the so-called link was the identification of metabolic pathways where gene of interests were involved in. Note: One way to integrate epigenomic data with transcriptomics and metabolomics would be to follow an approach similar to.16 Differentially Methylated Cytosines (DMCs) located within gene body regions can be used to identify a specific list of genes impacted by the trait of interest, identified by WGCNA. Because these genes retain the same KEGG gene identifiers as those used for transcriptomic data, this DMC-derived gene list can be mapped onto KEGG pathways and linked to compound IDs using the exact same integration strategy described in this protocol for the transcriptomics-metabolomics workflow. Troubleshooting Problem 1 Poor RNA-seq data quality due to low sequencing depth, RNA degradation, or contamination may reduce the reliability of gene expression estimates. Potential solution Perform strict quality control (e.g., read trimming, removal of low-quality reads) and verify mapping rates. Samples with poor sequencing metrics, qualified as outliers, should be excluded before downstream analyses. Problem 2 When computing a scale-free topology in an unsupervised signed network using WGCNA, it may occur that no soft-thresholding power yields a scale-free topology fit index (R2) above 0.8 (see Step 5). Although achieving a scale-free topology is recommended for network construction, this criterion is not always strictly met depending on the dataset. Potential solution If this occurs, it is advisable to check both the filtering of the raw data and the normalization step (Step 4). Excessive noise in the dataset, such as lowly expressed or highly variable genes, can reduce gene–gene correlations and prevent the network from approximating a scale-free topology. Problem 3 In WGCNA analysis (related to both individual omics analyses) the grey module should not be considered in downstream analyses. As discussed in Step 7, the grey module contains all genes that were not assigned to any co-expression module (e.g., genes showing low correlation with other genes in the dataset or belonging to clusters that are too small). Potential solution If more than 40–50% of genes are in the grey module, one should try to modify network characteristics (e.g., filter dataset, soft power threshold, minimal size of the modules). Problem 4 The WGCNA analysis (related to both individual omics analyses) may output no significantly correlated modules with the trait of interest (even when an appropriate soft-thresholding power has been selected). Potential solution Before modifying network parameters, verify that the dataset contains a sufficient number of samples and that data preprocessing steps (e.g., normalization and filtering) were appropriately performed. Increasing module sensitivity by adjusting parameters such as the minModuleSize or mergeCutHeight may help identify relevant modules. However, these adjustments should be applied cautiously, as forcing the detection of very small modules relative to the sample size may lead to results that are not biologically representative. Problem 5 Suspected batch effects, especially in the metabolomic analysis (related to Individual omics analysis: computational analysis of the metabolomics data). This can be due to samples being processed on different time points. This problem can be a major source of inconsistency during the analysis. Potential solution In MetaboAnalyst, you can fix this problem with. >mSet <- InitDataObjects(“conc”, “stat”, FALSE) >mSet <- Read.BatchDataTB(mSet, “#csv file containing the raw data”, “row”, “lods”) >mSet <- PerformBatchCorrection(mSet) After, you may want to confirm that samples cluster primarily according to biological conditions rather than technical factors. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Leo Duperret (duperretleo@gmail.com). Technical contact Technical questions on executing this protocol should be directed to and will be answered by the technical contact, Leo Duperret (duperretleo@gmail.com). Materials availability This study did not generate new unique reagents. Data and code availability RNA-seq data have been made available through the ENA database under Bioproject DECICOMP: PRJEB86647. RNA-seq data were submitted under accession numbers PRJEB86618 for the starved condition and PRJEB86525 for the fed condition. Metabolomic raw data were made publicly available in MetaboLights database (accession number MTBLS 13313). The code used for this article is available at https://github.com/lduperret/STAR_Protocol_Omics_integration. Acknowledgments We are grateful to Margot Doberva and Michèle Laudié from the Bio-Environment platform (UPVD, Région Occitanie, CPER 2007–2013 Technoviv, CPER 2015–2020 Technoviv2) for technical support in library preparation and sequencing. We thank the bioinformatics service of Ifremer (SEBIMER) for their help in bioinformatics. This study is set within the framework of the “Laboratoire d’Excellence (Labex)” TULIP (ANR-10-LABX-41). The present study was supported by the ANR project DECICOMP (ANR-19-CE20-0004) and the MAHEWA project (France 2030, ANR-23-POCE-0001). Author contributions B.P., J.d.L., J.V.-D., and G.M. designed the experiments. B.P., B.M., C.C., L. Degremont, N.F., G.C., J.V.-D., and G.M. performed oyster experiments. J.P., O.R., and G.C. extracted nuclear material. S.H. and A.T. performed metabolomic experiments. L. Duperret, A.V., E.T., S.H., A.T., J.V.-D., A.L., and G.M. performed RNA-seq and metabolomic analyses. L. Duperret, A.V., E.T., J.V.-D., A.L., and G.M. interpreted results and wrote the article. All the authors have revised and approved the manuscript submission. Declaration of interests The authors declare no competing interests. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.xpro.2026.104875. Supplemental information Table S1. List of metabolites identified for the targeted metabolomics analysis References Associated Data Supplementary Materials Table S1. List of metabolites identified for the targeted metabolomics analysis Data Availability Statement RNA-seq data have been made available through the ENA database under Bioproject DECICOMP: PRJEB86647. RNA-seq data were submitted under accession numbers PRJEB86618 for the starved condition and PRJEB86525 for the fed condition. Metabolomic raw data were made publicly available in MetaboLights database (accession number MTBLS 13313). The code used for this article is available at https://github.com/lduperret/STAR_Protocol_Omics_integration.