Skip to main content
v2026.11,986 entries · CC-BY 4.0

Building an Analytical Calibration Curve: Linear Range, R-squared and Back-Calculating a Concentration

How to build an analytical calibration curve from a standard set, read slope/intercept/R-squared and residuals, and back-calculate an unknown concentration, with a full worked least-squares example plus internal standard and standard addition methods.

Written and maintained by CASRAI Editorial Board

Published · Last updated

On this page

To back-calculate a concentration from a calibration curve, fit the line y = mx + b through the standards, then solve for x: concentration = (signal − intercept) / slope. A calibration curve (also called a standard curve) is the working relationship between the concentration of an analyte and the signal an instrument produces for it — absorbance, peak area, fluorescence intensity, or a similar response. You build it by measuring a set of standards of known concentration, fitting a line (usually by least-squares linear regression) through the resulting points, and then using that line to convert an unknown sample’s signal back into a concentration. Almost every quantitative method in a lab — UV-Vis spectrophotometry, HPLC, GC, ELISA, atomic absorption — depends on this same basic procedure underneath whatever the instrument-specific steps look like.

This guide walks through building a curve from a real five-point standard set, reading what the slope, intercept and R² actually tell you, and back-calculating a sample concentration from the fitted equation, plus the two methods (internal standard and standard addition) you reach for when a simple external calibration isn’t reliable.

Choosing Standards and the Linear Range

The linear range is the span of concentrations over which the instrument response is proportional to concentration — where a plot of response vs. concentration is a straight line within acceptable scatter. Every detector has an upper limit where the response flattens (detector saturation, self-absorption, antibody saturation in an immunoassay) and a lower limit set by noise and the limit of detection and limit of quantitation. Your standards need to bracket the concentration you expect to find in real samples, sitting inside that linear window rather than at its edges.

Decision General guidance
Number of standards At least five non-zero concentrations is the practical minimum for a routine curve; regulated bioanalytical work (ICH M10 and the FDA 2018 guidance) calls for a blank, a zero sample and at least six non-zero levels, and ICH Q2(R2) asks for at least five concentrations to show linearity. More points improve the precision of the fitted slope and intercept and make it easier to spot curvature.
Spacing Evenly spaced or log-spaced across the range, not clustered at one end. Clustering inflates the leverage of the few points at the other end and can make R² look better than the fit actually is across the working range.
Range width Bracket the expected sample concentration near the middle of the curve, not at an extreme. A sample reading near the top or bottom of the standard range carries more back-calculated error than one near the centroid.
Blank Run a true blank (matrix with no analyte) alongside the standards. It isn’t usually plotted as x=0 in the regression, but it tells you whether the intercept is behaving as expected and flags contamination.
Replicates Where practical, run each standard in duplicate or triplicate. It doesn’t change the regression math on the means, but it lets you see which points are noisy rather than trusting single readings.

Worked Example: Building the Curve by Least-Squares Regression

The table below is a five-point standard set for a colorimetric assay — concentration in mg/L on the x-axis, absorbance on the y-axis. The numbers are illustrative, built to show the arithmetic clearly rather than taken from a specific instrument run.

Standard Concentration, x (mg/L) Absorbance, y Fitted ŷ (from the line below) Residual (y − ŷ)
1 2 0.058 0.0590 −0.0010
2 4 0.112 0.1087 +0.0033
3 6 0.159 0.1584 +0.0006
4 8 0.201 0.2081 −0.0071
5 10 0.262 0.2578 +0.0042

Ordinary least-squares regression finds the slope (m) and intercept (b) of the line y = mx + b that minimizes the sum of squared vertical distances between each point and the line. With n = 5 points:

  • Σx = 30, Σy = 0.792, Σxy = 5.746, Σx² = 220
  • slope, m = (nΣxy − ΣxΣy) / (nΣx² − (Σx)²) = (5 × 5.746 − 30 × 0.792) / (5 × 220 − 900) = 0.02485
  • intercept, b = (Σy − mΣx) / n = (0.792 − 0.02485 × 30) / 5 = 0.00930

The fitted equation is y = 0.02485x + 0.00930. Plugging each standard’s x back into that equation gives the “fitted ŷ” column above; the differences between the measured y and fitted ŷ are the residuals, and it’s the sum of the squares of those residuals that the regression minimized.

Back-Calculated Accuracy of the Five Standards

Acceptance of a calibration curve in bioanalytical guidance rests on back-calculated standards, not on R². Each standard’s own signal is put through x = (y − b) / m and compared with its nominal concentration:

Standard Nominal x (mg/L) Signal y Back-calculated x (mg/L) Accuracy (% of nominal)
1 2 0.058 1.960 98.0
2 4 0.112 4.133 103.3
3 6 0.159 6.024 100.4
4 8 0.201 7.714 96.4
5 10 0.262 10.169 101.7

All five fall within ±15% of nominal, and the lowest standard (98.0%) would also satisfy the wider ±20% allowed at the lower limit of quantification. The largest deviation, standard 4 at 96.4%, corresponds to the largest residual (−0.0071) in the table above. The data are illustrative; the point is the procedure: compute the percentage for every standard before using the curve on unknowns.

Reading the Output: Slope, Intercept, R² and Residuals

Parameter What it tells you
Slope (sensitivity) How much the signal changes per unit of concentration. A shallow slope means low sensitivity — small concentration differences barely move the signal, which widens back-calculated error. A slope near zero across your whole range usually means you’ve exceeded the linear range or the method genuinely isn’t responsive there.
Intercept The predicted signal at zero concentration. A small non-zero intercept is normal (background absorbance, baseline noise). A large intercept relative to the signal from your lowest standard is a warning sign — contamination, an uncorrected blank, or a matrix effect.
R² (coefficient of determination) The fraction of variance in y explained by the linear fit with x. In the worked example, R² = 0.9968. It is a measure of correlation, not of accuracy or linearity by itself — a curved data set can still produce a deceptively high R², which is why it should never be the only thing you check (see the troubleshooting table below).
Residuals The point-by-point differences between measured and fitted signal. Randomly scattered residuals with no pattern support a linear model. Residuals that grow with concentration, or that curve systematically (all positive in the middle, negative at the ends, or vice versa), indicate the relationship isn’t actually linear across the range you’ve chosen.

What R² Is Held To, and Why That Is Not the Whole Story

R² ≥ 0.99 or ≥ 0.995 is a frequent convention in laboratory SOPs, and some methods specify a stricter value or the correlation coefficient r instead. These are local conventions, not guideline requirements: ICH Q2(R2) asks that linearity be shown with a plot of response against concentration, the correlation coefficient or R², the intercept, the slope and a residual-plot analysis, but it sets no numeric R² threshold, and ICH M10 judges a calibration curve by the back-calculated accuracy of its standards rather than by R². Treat any single R² figure as a starting point set by your method or SOP, and always look at the residual pattern alongside it. A high R² with a visibly curved residual plot means the linear range has been overstated; it is not a pass.

Linear Range, LOD/LOQ and Where the Curve Breaks Down

The calibration curve is only valid inside the concentration window you actually tested with standards. Below the lowest standard, you’re extrapolating into territory governed by the method’s limit of detection and limit of quantitation — the LOD is the lowest concentration reliably distinguishable from noise, and the LOQ (typically the lowest point you actually calibrate to) is the lowest concentration you can quantify with acceptable precision and accuracy. Above the highest standard, you risk detector saturation or self-absorption, where the true response curve bends over even though a straight line still fits reasonably well across your tested points. Never report a back-calculated concentration that falls outside the calibrated range without diluting (or concentrating) the sample and re-measuring inside the range — extrapolation past the standards is not the same as interpolation within them.

Back-Calculating an Unknown Concentration

Once the line is fitted, converting a sample’s measured signal into a concentration is a matter of solving the equation for x instead of y:

x = (y − b) / m

Using the worked example’s fitted equation (y = 0.02485x + 0.00930), suppose an unknown sample reads an absorbance of 0.150:

  • x = (0.150 − 0.00930) / 0.02485 = 5.662 mg/L

That 5.662 mg/L is the concentration in the diluted solution actually measured. If the sample was diluted 1:5 before measurement (a common step to bring a concentrated sample down into the linear range), the reported concentration in the original sample multiplies back through that dilution factor:

  • reported concentration = 5.662 mg/L × 5 = 28.31 mg/L

Two checks worth making before reporting that number: confirm 5.662 mg/L falls inside the tested standard range (2–10 mg/L here — it does, comfortably inside), and if you’re running quality-control samples of known concentration through the same curve, confirm the back-calculated QC value falls within your method’s accuracy window (commonly ±15% of nominal, tightened to ±20% only at the LOQ, in regulated bioanalytical practice) before trusting unknowns calculated from that curve.

When a Simple External Calibration Isn’t Enough: the Internal Standard Method

The worked example above is an external calibration — standards and unknowns are run under (ideally) identical conditions, and the curve built from the standards is applied directly to the unknowns’ raw signal. That works well when instrument response is stable run-to-run. It breaks down when injection volume, extraction recovery or instrument sensitivity drifts between the standard runs and the sample runs — common in chromatography and mass spectrometry, where small differences in injection volume or ionization efficiency between injections translate directly into concentration error.

The internal standard method corrects for that by adding a fixed, known amount of a second compound (the internal standard — chemically similar to the analyte, but resolvable from it) to every standard and every sample before analysis. Instead of plotting analyte signal vs. concentration, you plot the ratio of analyte signal to internal standard signal vs. concentration. Because both the analyte and the internal standard experience the same injection-volume or extraction variability in a given run, that variability cancels out in the ratio. This is standard practice in GC and LC-MS quantitation — see the internal-standard sections of the gas chromatography and LC-MS guides for how it’s applied downstream of the column.

Standard Addition Method

External calibration (with or without an internal standard) assumes the calibration standards and the real sample have the same matrix — that anything else in the sample besides the analyte doesn’t change how the instrument responds. When that assumption doesn’t hold (a complex biological matrix, a soil extract, a food sample with components that suppress or enhance the signal), the standard addition method avoids the problem by calibrating inside the actual sample matrix instead of a separate set of standards.

The procedure: take several aliquots of the same sample, spike each with a different, known additional amount of the analyte (leaving one aliquot unspiked), and measure all of them. Plotting signal vs. the added concentration gives a line whose x-intercept (found by extrapolating the fitted line back to y = 0) is the negative of the original, unknown concentration already present in the sample — because the matrix is identical across every aliquot, any matrix effect is baked equally into every point and cancels out of the extrapolation. The trade-off is that standard addition requires more sample and more measurements per unknown than a single shared external curve, so it’s typically reserved for matrices known or suspected to interfere, rather than used routinely.

What Regulators Require of a Calibration Curve

Two documents govern calibration for regulated bioanalysis (measuring drugs and their metabolites in biological samples): the ICH M10 guideline on bioanalytical method validation and study sample analysis (Step 4, 2022), and the FDA’s Bioanalytical Method Validation guidance for industry (May 2018). Their numbers are compared side by side in the data table below this guide. In outline:

  • One curve per analyte per run. ICH M10 section 3.2.4 expects a calibration curve for each analyte in each analytical run, prepared in the same biological matrix as the study samples.
  • Range. The calibration range runs from the lower limit of quantification (LLOQ, the lowest standard) to the upper limit of quantification (ULOQ, the highest standard).
  • Levels. A blank sample (matrix without analyte or internal standard), a zero sample (matrix with internal standard only) and at least six non-zero concentration levels covering the range, including the LLOQ and ULOQ. Blank and zero samples are not used in the regression; they show whether interference or carryover is present.
  • Accuracy of back-calculated standards (chromatographic methods). Within ±15% of nominal, except the LLOQ, which may deviate by up to ±20%. At least 75% of the standards, and a minimum of six levels, must meet this.
  • Accuracy of back-calculated standards (ligand-binding assays, such as immunoassays). Within ±20% of nominal, and ±25% at the LLOQ and ULOQ, again with at least 75% of standards and six levels passing.
  • Model. Use the simplest model that adequately describes the concentration–response relationship, with the regression model and weighting scheme fixed during validation; in-study runs use the same model.

These criteria apply to bioanalysis for drug development. A research assay outside that setting is not bound by them, but they are a defensible benchmark for any quantitative method that needs documented accuracy.

Rejecting a Calibration Standard

A standard that fails the accuracy criterion may be excluded, and the curve is then refitted without it, provided the remaining standards still meet the criteria. Only a failing point may be dropped, and dropping it must not change the regression model. If the excluded standard is the LLOQ or the ULOQ, the run’s quantifiable range shrinks to the lowest and highest standards that remain, and under ICH M10 the revised range must still cover at least three levels of quality-control samples. Results outside the surviving range cannot be reported from that run.

Linearity and Range in ICH Q2(R2)

ICH Q2(R2), validation of analytical procedures (2023), applies to procedures such as assay and impurity methods in pharmaceutical analysis. Its linearity expectations match what this guide has built: at least five concentrations across the range, a plot of response against concentration, the correlation coefficient or R², the slope and intercept, and a residual-plot analysis. It also distinguishes two ranges:

  • The working range is the interval between the lowest and highest concentrations presented to the instrument for which reliable results are obtained.
  • The reportable range is what the procedure can report. It may combine several working ranges, for example when over-range samples are diluted into the working range.

Q2(R2) also states that linearity is not a requirement for every response. Immunoassays and cell-based assays typically give S-shaped (sigmoidal) responses, and non-linear models such as four- or five-parameter logistic (4PL or 5PL) fits are accepted for them. FDA and ICH M10 take the same position for ligand-binding assays: the curve is fitted with the model that suits the response, and acceptance is judged by back-calculated accuracy. This matters for ELISA work, where forcing a straight line through a sigmoidal curve gives biased results at both ends.

Weighted Regression and Model Choice

Ordinary least squares assumes the scatter of the signal is the same at every concentration. In many methods, especially chromatography with mass spectrometric detection over a wide range, scatter grows with concentration (heteroscedasticity). An unweighted fit then lets the large standards dominate: the line is pulled toward the top of the range and the low standards, which matter most near the LLOQ, can show large percentage errors. Weighted regression gives each point a weight inversely related to its variance. The common choices are 1/x and 1/x² (and the same on the response y); the stronger the growth of variance with concentration, the stronger the weighting needed.

  • Choose the weighting from validation data, by comparing back-calculated accuracy across the range for unweighted, 1/x and 1/x² fits, not by whichever gives the highest R². Both ICH M10 and the FDA guidance say the regression model and weighting scheme are determined during validation and the simplest adequate model is used.
  • Use the same model and weighting in routine runs as in validation. Changing the weighting to rescue a failing run is not acceptable.
  • A plot of residuals against concentration shows the need: a funnel shape (small residuals at the bottom, large at the top) points to weighting; a systematic bow points to a non-linear model or a range that is too wide.
  • Back-calculation is unchanged: solve the weighted line for x. Only the slope and intercept differ.
  • Do not use a quadratic term just to improve the fit. A non-linear model needs a reason in the response chemistry, and the guidance prefers the simplest model that works.

Dilution Integrity When a Sample Is Above the Range

A sample above the ULOQ is diluted into the range, and the dilution itself must be shown to be reliable. ICH M10 (section 3.2.7) and the FDA guidance ask for dilution quality-control samples prepared above the ULOQ, diluted with blank matrix, with at least five replicates per dilution factor. The mean accuracy must be within ±15% and the precision (CV) no greater than 15%. The dilution factors validated are then the only ones that may be used on study samples. The arithmetic is the one shown earlier: the concentration read from the curve is multiplied by the dilution factor to give the concentration in the original sample, which means any error in the dilution is multiplied too. The serial dilution technique guide covers how to prepare dilutions accurately.

Standard Addition: Limits to Keep in Mind

The standard addition method described above assumes the signal is linear in concentration across the spiked range, because the unknown is found by extrapolating the line back to zero signal. Extrapolation carries more uncertainty than interpolation, so the unspiked sample should read well above the noise and the spikes should be large enough to roughly double the signal. As an editorial note on scope: standard addition is generally taken to correct matrix effects that change the slope of the response; a constant additive background from the sample is not corrected by it and needs a separate blank or background correction.

Troubleshooting: Common Calibration Curve Problems

Symptom Likely cause Fix
High R² but visibly curved residual plot Range extends into a non-linear region (detector saturation at the high end, or below the LOQ at the low end) Narrow the calibration range to the portion that’s genuinely linear, or switch to a quadratic/weighted fit only if your method validation supports a non-linear model
Low R² despite points looking roughly linear by eye One noisy or mis-pipetted standard is pulling the fit; or the range is too narrow relative to measurement noise, so scatter dominates Re-run the suspect standard; widen the range if noise-to-signal is the issue; check for a transcription or dilution error on that point
Large positive intercept Reagent blank not subtracted; contamination in the diluent or glassware; baseline drift Re-zero against a proper matrix blank; check glassware and reagent purity; re-run the blank
Slope much shallower than expected from prior runs Degraded standard stock, wrong stock concentration, or an instrument sensitivity drop (lamp aging in a spectrophotometer, detector fouling in a chromatograph) Verify the stock standard against a fresh reference material; check instrument performance — see the spectrophotometer calibration guide for a photometric-accuracy check
Unknown reads above the top standard Sample concentration genuinely exceeds the calibrated range Dilute the sample and re-measure inside the range; never extrapolate above the highest standard
Same nominal standard gives different results on different days Matrix effects, instrument drift, or a genuine change in the standard’s condition (degradation, evaporation concentrating the stock) Consider standard addition if matrix is the suspected cause; recalibrate each session rather than reusing an old curve; discard and remake aged stocks

Acceptance Criteria Summary

Check Typical target Notes
R² (or r) Often ≥ 0.99 to 0.995 in SOPs; no numeric value in ICH Q2(R2) or ICH M10 A local convention; always pair with a residual check and back-calculated accuracy
Residual pattern Random scatter, no systematic curve A patterned residual plot overrides a passing R²
Back-calculated standard accuracy Chromatographic: ±15% of nominal, ±20% at the LLOQ. Ligand-binding: ±20%, ±25% at the LLOQ and ULOQ (ICH M10, FDA 2018) Re-plug each standard’s own signal into the fitted equation and compare to its known concentration
Number of standards passing accuracy ICH M10 and FDA 2018: at least 75% of non-zero standards, and a minimum of six levels, must pass A failing standard at either end effectively shrinks your validated range

Frequently Asked Questions

What R² value is acceptable for a calibration curve?

No numeric R² is set by ICH Q2(R2) or ICH M10. Q2(R2) asks you to report R² or the correlation coefficient with the slope, intercept and a residual-plot analysis; M10 accepts a curve on the back-calculated accuracy of its standards. R² ≥ 0.99 to 0.995 is a common SOP convention, and some methods require ≥ 0.999. Check residuals and back-calculated accuracy as well, since a high R² does not rule out curvature.

How many standards do I need to build a calibration curve?

ICH Q2(R2) asks for at least five concentrations to show linearity. For bioanalytical methods, ICH M10 and the FDA 2018 guidance call for a blank, a zero sample (for chromatographic methods) and at least six non-zero levels including the LLOQ and ULOQ. More evenly spaced points make curvature easier to see.

What accuracy must back-calculated standards meet?

For chromatographic methods, within ±15% of nominal (±20% at the LLOQ). For ligand-binding assays, within ±20% (±25% at the LLOQ and ULOQ). In both cases at least 75% of the standards, and at least six levels, must meet the limit.

When should I use weighted regression?

When the scatter of the response grows with concentration, so that residuals form a funnel. Pick the weighting (typically 1/x or 1/x²) during validation by comparing back-calculated accuracy across the range, and keep the same weighting in routine use.

Can I reject a calibration standard?

A standard failing the accuracy criterion can be excluded and the curve refitted, if the remaining standards still pass and the model is unchanged. If the LLOQ or ULOQ standard is excluded, the usable range of that run becomes shorter.

What is the difference between linear range, working range and reportable range?

The linear range is where response is proportional to concentration. The working range, in ICH Q2(R2), is the interval between the lowest and highest concentrations presented to the instrument that give reliable results. The reportable range can combine several working ranges through dilution.

How do I handle a sample above the ULOQ?

Dilute it into the range using a dilution factor that was validated with dilution quality-control samples (at least five replicates, accuracy within ±15%, CV ≤ 15% under ICH M10), then multiply the back-calculated result by the dilution factor.

Can I back-calculate a concentration that falls outside my calibration range?

Not reliably. The fitted line is only validated across the concentration window actually bracketed by your standards. A signal outside that window should be brought back into range by diluting or concentrating the sample and re-measuring, not by extrapolating the equation past the tested points.

What’s the difference between the internal standard method and the standard addition method?

The internal standard method adds a fixed known amount of a second, chemically similar compound to every standard and sample to correct for run-to-run variability in injection volume or instrument response; it still uses a separate standard curve. The standard addition method spikes known amounts of the analyte itself directly into aliquots of the real sample, calibrating inside the sample’s own matrix to correct for matrix effects that a separate standard curve can’t capture.

Is a calibration curve the same thing as a standard curve?

Yes — the terms are used interchangeably in most analytical chemistry contexts, both describing the fitted relationship between known standard concentrations and instrument response used to quantify unknowns.

Sources

Sources:

  • ICH M10, Bioanalytical Method Validation and Study Sample Analysis, Step 4 (2022): database.ich.org. Sections 3.2.4 (calibration curve), 3.2.7 (dilution integrity) and 4.3.2 (run acceptance).
  • FDA, Bioanalytical Method Validation Guidance for Industry (May 2018): fda.gov.
  • ICH Q2(R2), Validation of Analytical Procedures (2023): database.ich.org. Sections 3.2.1 (range) and 3.2.2 (linearity).
  • Worked example arithmetic: least-squares calculation on the five illustrative standards above.

For the instrument-specific calibration checks that keep the hardware itself producing a trustworthy signal in the first place, see spectrophotometer calibration, pH meter calibration, analytical balance calibration and the microplate reader calibration guide. For the underlying accuracy/precision distinction that residual scatter is really measuring, see pipetting technique: accuracy and precision. For methods where a calibration curve is the core quantitation step, see ELISA protocol basics, UV-Vis spectrophotometer basics, HPLC and atomic absorption spectroscopy. For preparing the standards themselves, see serial dilution technique and molarity and solution calculations. For what “calibrated” means on the certificate an instrument or reference standard ships with, see calibration certificates and metrological traceability.

Data: Calibration curve requirements compared across ICH M10, FDA 2018 BMV and ICH Q2(R2)

Calibration-curve design and acceptance requirements for chromatographic and ligand-binding bioanalytical methods and for analytical procedure validation, compiled by reading the ICH M10 guideline (Step 4, 2022), the FDA Bioanalytical Method Validation guidance (May 2018, Table 1) and ICH Q2(R2) (2023) and aligning their stated numbers side by side.

Period covered: 2018/2022.

Calibration curve requirements compared across ICH M10, FDA 2018 BMV and ICH Q2(R2)
Method type and guidelineMinimum non-zero calibration levels (levels)Blank and zero samples requiredBack-calculated accuracy, standard levels (% of nominal)Back-calculated accuracy, LLOQ (% of nominal)Back-calculated accuracy, ULOQ (% of nominal)Share of standards that must pass (%)
Chromatographic, ICH M10 (section 3.2.4)6Blank and zero sample15201575
Chromatographic, FDA 2018 (Table 1)6Blank and zero calibrator15201575
Ligand-binding, ICH M10 (section 3.2.4)6Blank and zero sample20252575
Ligand-binding, FDA 2018 (Table 1)6Blank (zero only at in-study analysis)20252575

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · full access

Ask about Building an Analytical Calibration Curve: Linear Range, R-squared and Back-Calculating a Concentration

Full access: large monthly usage, up to 10 source passages behind each answer, and MCP and API access for Claude Code, Codex and other CLIs. Cancel any time.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →