Methods & Caveats

A plain-language summary of how the datasets on this site were produced, and the limitations that must frame every interpretation. See the analysis repository for full code and parameters.

The single most important caveat. Every interaction on this site is computationally predicted and represents a hypothesis for future experimental testing. None have been experimentally validated. Predicted binding does not establish functional targeting, and correlation does not establish causation.

How the data were produced

RNA-seq (mRNA) processing: Reads were quality-trimmed (fastp), aligned to each species genome (HISAT2), and assembled (StringTie); a gene count matrix was produced with prepDE.py.
lncRNA discovery: Assembled transcripts were compared to annotations (gffcompare), size-filtered (>199 bp), and screened for coding potential (CPC2); non-coding transcripts ≥5 reads in ≥1 sample were retained and counted (featureCounts).
miRNA discovery: Small RNA-seq reads were processed with ShortStack (de novo + a miRBase/cnidarian reference), and named following Ashey et al. (2026). Known families keep their names; unmatched sequences are labelled -mir-novel-#.
Whole-genome bisulfite sequencing (WGBS): Reads were trimmed (cutadapt), aligned (Bismark), deduplicated, and methylation extracted; CpGs were filtered to ≥10× coverage in ≥4 of 5 samples (methylKit) before feature and expression analyses.
miRNA binding prediction: miRanda (v3.3) predicted binding to CDS, 5′UTR, and 3′UTR (UTRs defined as 1 kb up/downstream), with strict seed complementarity, score ≥ 100, and energy ≤ −20 kcal/mol. Cnidarian miRNAs bind with near-full complementarity (plant-like), so CDS binding is biologically relevant.
Coexpression analysis: Pairwise Pearson correlation was computed on RPM-normalized counts for each predicted binding pair (and for all lncRNA–mRNA pairs). Significance used an unadjusted p<0.05.
ceRNA inference: Candidate ceRNAs are lncRNAs predicted to bind a miRNA, negatively coexpressed with that miRNA, and positively coexpressed with the miRNA's mRNA target — a signature consistent with sequestration and indirect derepression, but far from the only interpretation. Only a small minority of the lncRNAs identified in each species met these criteria; the great majority were not classified as ceRNA candidates. ceRNA sponging is one of many documented lncRNA functions (e.g. chromatin scaffolding, transcriptional and splicing regulation, decoys for proteins), most of which are not evaluated here.
Epigenetic machinery & epi-miRNAs: Epigenetic-machinery homologs were identified (blastp against curated references). miRNAs predicted to target these transcripts are reported as putative epi-miRNAs.
Functional enrichment: Target genes were annotated with GO terms (UniProt/SwissProt) and tested for over-representation with topGO (weight01, Fisher).

Limitations you must keep in mind

Five biological replicates per species. All correlations rest on n = 5.
Pairwise correlations are sensitive to small sample size. With n = 5, a handful of points drive each coefficient.
Unadjusted p-values. Interactions were prioritized by correlation magnitude (|r| ≈ 0.88 at p<0.05, n = 5), not multiple-testing-corrected significance — expect false positives.
Predicted binding ≠ functional targeting. miRanda predicts thermodynamically plausible duplexes, not validated regulation.
Positive coexpression is common and ambiguous. It may reflect co-regulation, indirect loops, or non-canonical effects rather than repression.
ceRNA relationships are inferred, not validated. They are consistent with sponge behaviour but not demonstrated experimentally.
ceRNA is one of several possible lncRNA functions. Only a small fraction of the lncRNAs identified were predicted to act as ceRNAs; the majority were not part of any ceRNA network, and their functions — which need not involve miRNA sponging at all — remain uncharacterized. The ceRNA framing used across this site reflects the questions asked, not the full functional repertoire of these transcripts.
Genome assembly quality varies among species and influences target prediction; cross-species comparisons should be interpreted cautiously.
Network edges are hypotheses for future experimental testing, not established mechanisms.

Language we use (and avoid)

We say “predicted to target”, “putatively interacts”, “consistent with”, “candidate ceRNA”, and “hypothesized regulatory relationship”. We avoid “proves”, “controls”, or “causes” unless directly supported by experiment — which, here, they are not.

Statistics and framing on this page are transcribed from the manuscript "Cross-talk among miRNAs, lncRNAs, and DNA methylation in three coral species reveal conserved epigenetic regulatory architecture."