RNA Integrity Number (RIN): How It Is Computed and How to Read a Trace

The RNA Integrity Number (RIN) is a single value from 1 to 10 that an automated electrophoresis instrument assigns to a total RNA sample, where 10 is fully intact and 1 is totally degraded. It is the number a sequencing core quotes back at you when it accepts or rejects a submission, and the number a reviewer expects to see in a methods section. It is also one of the most widely misdescribed metrics in molecular biology: it is routinely explained as the 28S:18S ribosomal ratio, or as a refinement of it. It is neither. RIN is the output of a trained model that reads the entire electropherogram, and it exists specifically because the ribosomal ratio did not work.

This page covers what the algorithm actually computes, what value you need for which downstream application and why that depends on target length rather than on a universal cut-off, how to read a degraded or atypical trace, and the sample types for which the number does not apply at all.

RIN is not the 28S:18S ratio — start here

Before RIN, RNA integrity was judged by the ratio of the areas (or heights) of the 28S and 18S ribosomal RNA peaks, with a value near 2.0 taken as the mark of intact eukaryotic total RNA. That convention is still quoted in protocols and course notes as though it were equivalent to a RIN. The paper that introduced RIN — Schroeder et al., BMC Molecular Biology 2006 — was written to replace it, and reported directly why.

The authors found that “the usage of ribosomal ratio for RNA quality assessment has several shortcomings” and that “in many cases, ribosomal ratios showed only weak correlation with RNA integrity.” Tested against actual downstream performance — real-time PCR measurement of housekeeping transcripts — the ribosomal ratio correlated with observed expression at 0.24, while RIN correlated at 0.52. Most concretely of all, they reported that applying “the historical value of 2.0 would reject about 40 experiments of good quality.”

Two practical consequences follow, and they run in both directions:

• A ratio well below 2.0 is not evidence that a sample is degraded. Intact RNA from many tissues and species simply does not present a 2:1 ratio, and the ratio is sensitive to how the peaks are integrated.
• A ratio at or near 2.0 is not evidence that a sample is intact. The ratio is computed from two peaks and is blind to everything between and below them — which is exactly where early degradation shows up first.

If a protocol, a core facility form or a methods section treats a ribosomal ratio and a RIN as the same claim, that is an error worth correcting rather than a stylistic preference.

How the RIN algorithm actually works

The algorithm does not look at two peaks. It segments the whole electropherogram into nine regions — the pre-region, marker region, 5S region, fast region, 18S region, inter-region, 28S region, precursor region and post-region — and computes candidate features across all of them. The “fast region” is the stretch between the alignment marker and the 18S peak; it is where short degradation fragments accumulate, and it carries a great deal of the model’s weight.

From a large pool of candidate features, forward selection by mutual information ranked the informative ones. In order of the information each contributed, the selected features were:

1. the total RNA ratio (the ratio of the ribosomal area to the total area of the trace);
2. the height of the 28S peak;
3. the fast-region area ratio;
4. the ratio of the combined 18S and 28S area to the fast-region area;
5. a linear regression of the fast-region endpoint;
6. the amount of detected fragment material in the fast region;
7. the presence or absence of the 18S peak;
8. a comparison of the overall mean value of the trace to its median.

Those features were then fed to an artificial neural network trained with Bayesian learning. Models were compared on evidence, and the published model uses five features and four hidden neurons. It was trained on 1,208 total RNA samples — 937 in the training set and 402 held out for testing — drawn from human, mouse and rat liver, kidney, colon, spleen, brain, heart and placenta, deliberately spanning intact through almost completely degraded material at different concentrations and from different extraction methods. Ten categories were defined, from 1 for totally degraded RNA to 10 for fully intact RNA.

The motivation for building a model at all is worth knowing, because it explains what the number is for. Human experts classifying the same traces disagreed at category boundaries: as the authors put it, “especially for RNA samples at the borderline of two adjacent integrity categories, the assignment to each of the two categories could be justified but one had to be selected.” RIN was built to make that assignment reproducible and automatic across labs and instruments, not to add information a careful person reading the trace could not see.

The single most useful thing to take from the algorithm’s structure: a sample can present textbook 18S and 28S peaks and still score in the middle of the scale, because a rising baseline of short fragments in the fast region moves several of the top-ranked features at once. That is not a software fault. It is the metric doing the job the ribosomal ratio could not.

What the number measures — and the caveat the reporting standard itself attaches

RIN is computed from ribosomal RNA landmarks, and rRNA is the overwhelming majority of a total RNA prep. Your experiment, in almost every case, is about the messenger RNA. The relationship between the two is a good proxy, not an identity, and the MIQE guidelines for qPCR reporting say so explicitly. On the use of a Bioanalyzer-style instrument to compute an integrity number, MIQE states that the advantage is that “these measures provide quantitative information about the general state of the RNA sample” but that “it is important to bear in mind, however, that these numbers relate to rRNA quality and cannot be expected to be an absolute measure of quality.”

MIQE also notes that mRNA degradation is not uniform: “even high-quality RNA samples can show differential degradation of individual mRNAs,” because mRNA turnover is a regulated biological process and not only a handling artefact. A high RIN says the sample was handled well. It does not certify that your particular transcript survived.

Where that distinction matters, MIQE points at a transcript-level alternative rather than a better rRNA metric: a reference gene/target gene 3′:5′ integrity assay, in which two amplicons are placed at opposite ends of the same transcript and their quantification cycles compared. MIQE states that the assay requires the PCR efficiencies of both assays to be virtually identical, that it needs an explicit threshold criterion, and that it should ideally target a panel of integrity reference genes “with a 3′:5′ threshold ratio of approximately 0.25.” It is a direct measure of the thing RIN only approximates.

One more thing RIN is not: a purity measurement. A260/A280 and A260/A230 tell you about protein, phenol and chaotrope carry-over; a spectrophotometer or fluorometer tells you concentration. Those are separate checks on separate instruments — see Qubit vs NanoDrop for which one answers which question. A sample can be pure, correctly quantified and thoroughly degraded.

RIN, RIN^e, RQN and DV₂₀₀ are four different numbers

These are routinely written as if they were the same measurement on different machines. They are not, and a methods section that reports “RIN > 7” for samples run on a TapeStation is reporting a number the instrument did not produce.

DIN (DNA Integrity Number) is the DNA analogue and has nothing to do with RNA. For the instrument-level differences behind these scores — chip versus cassette, throughput, sample number and hands-on time — see Bioanalyzer vs. TapeStation.

The reporting rule that follows: record the metric name, the instrument, the assay kit and the software version, and never convert one score into another. A conversion table between RIN and RIN^e would require the two models to be calibrated against a shared reference; they are not.

What value do you need? Ask how long a fragment your assay must recover

There is no universal acceptance threshold, and the reason is mechanical rather than conventional. Degradation is approximately random breakage along the length of a molecule. The probability that a given stretch of transcript survives without a break falls as that stretch gets longer. So the question that determines whether a degraded sample is usable is not “what is the RIN” but “what is the longest contiguous piece my protocol has to recover in one molecule?”

That single principle explains the whole pattern of published thresholds: an assay reading a 90 bp amplicon and an assay reconstructing a 3 kb full-length transcript are not the same request of the same material.

These are conventions, not standards. No standards body sets a RIN acceptance threshold. The authority for your experiment is your kit’s protocol and your core facility’s submission specification, and they differ. Treat a published number as a starting expectation to check, and record the one you actually applied. For how the integrity result feeds the poly-A versus rRNA-depletion decision in a sequencing experiment, see RNA-seq: experimental design through analysis; for the kit-level consequences, NGS library prep kits.

One consequence is worth stating plainly because it is where good samples get thrown away: a mid-scale RIN is a reason to change the protocol, not automatically a reason to discard the sample. A RIN that fails a poly-A library prep may be entirely adequate for a short-amplicon RT-qPCR panel, and the same sample can support both conclusions honestly as long as the integrity value is reported alongside the result.

FFPE and DV₂₀₀: a different measure for a different failure mode

Formalin-fixed, paraffin-embedded tissue is the case where the RIN scale actively misleads. Fixation crosslinks nucleic acid to protein and fragments the RNA chemically, and the ribosomal landmarks the algorithm keys on are largely gone. The trace produces a very low RIN, or the software declines to assign one — while the material can still yield a usable library.

DV₂₀₀ answers the question that library construction actually asks: what proportion of this material is long enough to be a template? It is the percentage of fragments longer than 200 nucleotides, read directly off the same trace, with roughly 30 % commonly cited as a practical working minimum for FFPE library prep. Because it is a percentage of a size distribution rather than a model score, it degrades gracefully where RIN falls off a cliff.

The practical rule: for FFPE, report DV₂₀₀ and do not treat the low RIN as a rejection criterion. A pathology archive sample with a RIN of 2 and a DV₂₀₀ of 45 % is a different proposition from a fresh-frozen sample with a RIN of 2, which is a handling failure.

Reading the trace: what each shape means

The trace carries information the score compresses away, which is why MIQE lists electropherogram traces as reportable material in their own right. What to look for, on a eukaryotic total RNA assay:

If you need a coarse orthogonal check without an automated electrophoresis instrument, a denaturing agarose gel showing two sharp ribosomal bands is the low-resolution version of the same read. MIQE’s position is that this is the floor, not the goal: it states that gel electrophoresis evidence is what one should provide “at the least or, better yet, results from a microfluidics-based rRNA analysis or a reference gene/target gene 3′:5′ integrity assay.”

Where RIN does not apply — the traps

The algorithm keys on eukaryotic ribosomal landmarks. Several common sample types do not present them, and the resulting number is either absent, or present and meaningless. This is the failure mode most likely to send a good sample to the bin.

Prokaryotic RNA

The frequently repeated claim that “RIN is not defined for bacterial RNA” is close to the truth but not quite it, and the distinction matters. The published RIN model was developed entirely on eukaryotic data, using the 5S, 18S and 28S peaks. Agilent’s 2100 Expert software, however, ships a separate prokaryote total RNA assay setting with its own calculation keyed on the bacterial ribosomal landmarks. As documented by Salazar et al. (2013) working with recombinant E. coli, the prokaryotic calculation weights the total RNA ratio, the 23S peak height, the 23S area ratio, a comparison of the 16S and 23S area to the fast-region area, a linear regression of the fast-region endpoint, detected fragment amounts in the fast region, the presence or absence of the 16S peak, and a comparison of the overall mean to the median — the same shape of model against 16S and 23S rather than 18S and 28S. They further note that for E. coli “a 5S peak is not detected, and thus not used in the calculation.”

So a prokaryotic integrity score does exist. What follows for a microbiology lab is narrower and more practical than the blanket claim:

• Select the prokaryote assay. Run bacterial RNA under the eukaryote assay and the algorithm is searching for 18S and 28S peaks that are not there.
• Do not compare a prokaryotic score with a eukaryotic one, or pool them in a table. They come from different models against different landmarks, and the prokaryotic one is not the score Schroeder et al. validated against downstream expression.
• Report the assay, not just the number. “RIN 8.2” is ambiguous for a bacterial sample; “RIN 8.2, prokaryote total RNA assay” is not.

Insects and other invertebrates with a hidden break

The 28S rRNA of most insects carries an endogenous “hidden break”. The two halves are held together by hydrogen bonding, so denaturation — the standard sample preparation for this assay — dissociates them into two fragments of similar size that migrate with the 18S peak. Perfectly intact insect RNA therefore presents as a single major peak where the algorithm expects two, and scores accordingly.

An application note by Fabrick and Hull, published by Agilent in cooperation with USDA-ARS, surveyed a range of insect species on the 2100 Bioanalyzer and found that “because of the eukaryotic total RNA algorithm used and the relative positioning of putative 18S, 28S, and other minor peaks found in RNA from different insects, RINs were not always automatically calculated” — the software returned no value at all for several species. Their conclusion is stated plainly: “the validity of RIN for insect total RNA was not proven,” and although “RIN cannot be reliably used for insect RNA, the electropherogram is a useful tool for the assessment of RNA degradation.” The same note records that where samples are not heat-denatured, the expected 18S and 28S peaks are observed.

The general lesson extends beyond insects: a missing RIN is a signal, not an error. When the software declines to assign one, the usual reason is that the trace does not match the model the assay assumes.

Anything without ribosomal peaks

• Purified or enriched RNA — poly-A-selected mRNA, rRNA-depleted RNA, size-selected small RNA, in-vitro transcripts. By design there is little or no rRNA left, so the algorithm has nothing to read. Assess integrity on the total RNA before enrichment and record it there.
• Plant material — chloroplast and mitochondrial rRNA add peaks to the eukaryotic pattern. Use the plant-specific assay.
• Mixed communities — host-microbiome, symbiont and metagenomic samples carry both prokaryotic and eukaryotic peak sets in one trace, and neither assay’s model describes it.
• Degraded archival material — see the FFPE section above; use DV₂₀₀.

What to report

Integrity reporting is not optional in qPCR work. The 2009 MIQE checklist listed, as essential, both “RNA integrity: method/instrument” and “RIN/RQI or Cq of 3′ and 5′ transcripts”, with electrophoresis traces listed as desirable. MIQE also states that “if the RNA sample is shown to be partially degraded, it is essential that this information be reported, because the assay’s sensitivity for detecting a low-level transcript may be reduced and relative differences in the degradation of transcripts may produce incorrect target ratios.” Degradation is not only a sensitivity problem; because transcripts degrade at different rates, it can move a ratio between targets in a direction that looks like biology.

Note that the essential/desirable tiering is historical. MIQE 2.0 (2025) replaced it with a single unified Yes/No checklist across five sections, with footnotes marking highly desirable and conditionally applicable items. If you are citing MIQE in a methods section written now, cite the 2.0 revision rather than the 2009 paper, and do not describe the current checklist as an E/D tiered list.

A minimum reportable set, whatever the guideline version:

1. The metric name as the instrument produced it — RIN, RIN^e, RQN or DV₂₀₀.
2. The instrument, assay kit and software version. The score is a model output and the model is versioned.
3. The value for every sample, not a range or a group mean. A mean RIN across a cohort conceals exactly the between-sample variation that becomes a batch effect.
4. The traces, as supplementary material.
5. Any sample retained below your stated threshold, and the justification for retaining it.

The last point is the one that most often separates a defensible methods section from an unreproducible one. Integrity varies systematically with collection site, ischaemic time and storage, so if it also varies systematically with your experimental groups, it is a confounder and belongs in the analysis, not only in the QC table.

Related pages

• RNA extraction protocol basics — where integrity is won or lost, and where genomic DNA carry-through starts.
• RNA-seq: experimental design through analysis — how the integrity result drives the poly-A versus rRNA-depletion decision and the rest of the study design.
• Bioanalyzer vs. TapeStation — the instrument-level comparison behind RIN, RIN^e and RQN.
• Qubit vs NanoDrop — the separate concentration and purity checks that integrity does not cover.
• Agarose gel electrophoresis protocol basics — the low-resolution orthogonal integrity check.
• qPCR melt curve analysis — the specificity check on the RT-qPCR assays that tolerate degraded input.
• NGS library prep kits and NGS service procurement — where integrity thresholds appear as acceptance criteria in a specification.
• Mycoplasma testing methods compared — the other routine gate to clear before committing cultured material to an expensive run.
• Mass spectrometry proteomics: DDA, DIA and run QC — the analogous sample-integrity and run-QC discipline on the protein side of a multi-omic study.
• Sanger sequencing — reading a different kind of trace, where peak shape is also the diagnostic.
• Laboratory equipment and instrumentation — the wider cluster this page belongs to.

Frequently asked questions

What is a good RIN number?

There is no single answer, because the acceptable value depends on the longest contiguous fragment your protocol has to recover. A RIN of roughly 7 or above is the most commonly cited expectation for standard poly-A RNA-seq; full-length and isoform-resolved protocols are stricter; short-amplicon RT-qPCR tolerates considerably less intact material. No standards body sets a threshold — your kit protocol and your core facility’s submission specification are the authority, and you should record the threshold you actually applied.

Is RIN the same as the 28S:18S ratio?

No, and this is the most common misconception about the metric. RIN was introduced in 2006 specifically to replace the ribosomal ratio, whose authors reported it showed only weak correlation with integrity — 0.24 against observed expression, versus 0.52 for RIN — and that applying the traditional 2.0 cut-off would have rejected about 40 samples of good quality in their own dataset. RIN is a model output over nine regions of the whole electropherogram; the ratio uses two peaks.

What does a RIN of 10 mean?

It means the algorithm placed the trace in the top of ten categories defined from 1 (totally degraded) to 10 (fully intact). It is a classification of the ribosomal profile, not a certification that any particular transcript is undamaged — MIQE’s own caveat is that these numbers “relate to rRNA quality and cannot be expected to be an absolute measure of quality.”

Can I use RNA with a RIN of 5?

For a short-amplicon RT-qPCR assay, frequently yes. For a poly-A-selected sequencing library, usually not without switching to an rRNA-depletion protocol, because poly-A capture compounds degradation with 3′ bias. The decision is about target length, not about the number in isolation — and whichever way it goes, report the value and the reasoning.

What is the difference between RIN and DV₂₀₀?

RIN is a 1–10 model score derived from ribosomal peak structure. DV₂₀₀ is a percentage: the proportion of fragments longer than 200 nucleotides, integrated directly from the size distribution. They are different units answering different questions and cannot be converted. DV₂₀₀ is the metric that stays informative for FFPE material, where the ribosomal landmarks RIN needs have been destroyed by fixation.

Is RIN^e the same as RIN?

No. RIN^e is the TapeStation’s RIN-equivalent score: the same 1–10 scale, computed differently on a different separation format. Report the metric the instrument actually produced rather than relabelling it, and do not attempt to convert between the two.

Why did my Bioanalyzer not return a RIN?

Because the trace did not match the model the assay assumes. Common causes: the sample is outside the assay’s quantitative concentration range; the wrong assay was selected for the sample type; the material is not eukaryotic total RNA (enriched, prokaryotic, plant or mixed); or the ribosomal peaks are absent or displaced, as with insect RNA and FFPE. A missing value is diagnostic information, not a software failure — work out which of those applies before re-running.

Does RIN apply to bacterial RNA?

The published RIN model was trained entirely on eukaryotic total RNA using the 5S, 18S and 28S peaks, so it does not. Agilent’s software does provide a separate prokaryote total RNA assay whose calculation is keyed on 16S and 23S instead, and it will return a score — but it is a different model against different landmarks, so a prokaryotic score should not be compared with a eukaryotic one or reported without naming the assay.

Do I need to report RNA integrity in a publication?

For qPCR work, yes: the 2009 MIQE checklist made both the integrity method/instrument and the integrity value essential reportable items, and MIQE 2.0 (2025) carries the requirement forward in its restructured checklist. MIQE is explicit that partial degradation must be reported, because it reduces sensitivity for low-abundance transcripts and can distort ratios between targets.

A note on sources

The algorithm description, the training-set composition and the ribosomal-ratio correlation figures on this page are taken from the primary paper: Schroeder A, Mueller O, Stocker S, Salowsky R, Leiber M, Gassmann M, Lightfoot S, Menzel W, Granzow M, Ragg T. “The RIN: an RNA integrity number for assigning integrity values to RNA measurements.” BMC Molecular Biology 2006;7:3. doi:10.1186/1471-2199-7-3. Quoted reporting requirements and the rRNA caveat are from Bustin SA, Benes V, Garson JA, et al. “The MIQE Guidelines.” Clinical Chemistry 2009;55(4):611–622. doi:10.1373/clinchem.2008.112797. The current revision is Bustin SA, Ruijter JM, van den Hoff MJB, et al. “MIQE 2.0.” Clinical Chemistry 2025;71(6):634–651. doi:10.1093/clinchem/hvaf043; its full checklist text is paywalled, so this page cites its structure rather than item numbers. The prokaryotic feature list is from Salazar MA, Fernando LP, Baig F, Harcum SW. “The effects of protein solubility on the RNA Integrity Number (RIN) for recombinant Escherichia coli.” Biochemical Engineering Journal 2013;79:129–135. doi:10.1016/j.bej.2013.07.011. The insect findings are from Fabrick JA and Hull JJ, “Assessing Integrity of Insect RNA,” Agilent Technologies application note 5991-7903EN (2017), a cooperative investigation with USDA-ARS.

On vendor documentation: agilent.com returned HTTP 403 to direct retrieval when this page was written, so no page on that domain is cited here. The single vendor-authored document used is the application note above, retrieved from the USDA-ARS copy. Instrument behaviour, assay ranges and software versions change; the acceptance thresholds described here are widely used conventions rather than standards, and your own kit protocol, instrument documentation and core facility specification are the authority for the values you apply.