A Bayesian hierarchical modeling approach can improve measurement accuracy of microcystin concentrations

Bayesian hierarchical modeling improves measurement accuracy of microcystin concentrations

Significance of the topic

Accurate measurement of microcystins (MCs) is critical for public health, ecosystem protection and water-management decisions during cyanobacterial harmful algal bloom (HAB) events. Calibration-based assays such as ELISA are widely used for routine monitoring but often rely on few standard points per plate, producing high uncertainty in estimated concentrations that can trigger costly management responses or miss hazardous exposures. The Bayesian hierarchical model (BHM) provides a statistical framework to reduce calibration uncertainty by borrowing strength across samples within a test and across historical tests, improving robustness without requiring changes to laboratory protocols.

Objectives and study overview

The study evaluates the effectiveness and operational feasibility of a BHM approach, including a sequential updating algorithm, to improve the accuracy of ELISA-derived microcystin concentration estimates. The authors apply the method to 214 ELISA test results collected by NOAA-GLERL from Lake Erie (2012–2021) and compare BHM variants to standard inverse-function calibration procedures and non-hierarchical Bayesian estimation.

Methodology

• Dataset: 214 ELISA test kits, each containing six standards (0–5.00 µg/L) and one control QC sample with known MC = 0.75 µg/L; up to 40 unknown samples per test, measured in duplicate.
• Calibration curves: earlier tests used a log–log linear approximation; later tests used a four-parameter logistic (4PL) nonlinear model. Small numbers of standards (five non-zero points after relative absorbance transformation) produce very low degrees of freedom (df ≤ 3), complicating frequentist uncertainty quantification.
• Bayesian hierarchical framework: two hierarchical levels were used — (1) within-test sharing among unknown sample concentrations (shrinkage toward the test-level mean) and (2) across-test sharing of calibration-curve parameters (hyperpriors for curve coefficients). Weakly informative priors were used initially; posterior summaries from an initial pool of nine tests were used to form informative priors for sequential updating.
• Sequential updating: implemented via MCMC (Stan through R). Posterior samples of hyperparameters (mean and variance of curve coefficients) were summarized using a normal–inverse-gamma representation and then used as conjugate priors for the next test. This allowed one-test-at-a-time analysis while accumulating information across tests without refitting the full historical dataset each time.
• Compared models: six approaches were evaluated — (a) IFE5: inverse-function estimator using 5 effective standard points (current practice); (b) IFE12: inverse-function with both replicates for all six standards (12 points); (c) Bayes: non-hierarchical Bayesian estimator with 12 points; (d) BHM1: hierarchical sharing within a test only; (e) BHM2: hierarchical sharing across tests only (with sequential updating); (f) BHM3: both within- and across-test hierarchical sharing (with sequential updating).
• Evaluation metric: accuracy defined as the absolute difference between the estimated QC concentration (posterior samples or Monte Carlo draws) and the known QC value (0.75 µg/L); medians and credible intervals summarized across tests.

Instrumentation and computational tools

• ELISA assay: Abraxis MC ELISA kits (six standards, QC sample per kit).
• Laboratory practice: duplicate measurements per unknown sample, average of replicates used.
• Computational tools: R and Stan for Bayesian modeling and MCMC sampling; sequential updating implemented by summarizing posterior hyperparameters and using conjugate prior forms (normal–inverse-gamma) for subsequent tests.

Main results and discussion

• Using both replicates for all six standards (IFE12) substantially improved accuracy relative to the common 5-point approach (IFE5): median accuracy improved from 0.261 µg/L (IFE5) to 0.145 µg/L (IFE12); variance of inaccuracies was significantly reduced (F-test p ≪ 0.001).
• Non-hierarchical Bayesian estimation (Bayes, with 12 standards) further reduced median inaccuracy to 0.127 µg/L by regularizing extreme estimates via weak priors.
• BHM variants that introduce shrinkage produce the best performance: BHM1 (within-test) median accuracy = 0.114 µg/L; BHM2 (across-test with sequential updating) = 0.121 µg/L; BHM3 (within- and across-test) = 0.109 µg/L — the best overall. The gains were modest but consistent, with narrower error distributions than inverse-function methods.
• The relatively small additional benefit of across-test pooling (BHM3 vs BHM1) was attributed to substantial heterogeneity among calibration curves: high across-test variance downweights the influence of pooling and makes within-test shrinkage the dominant advantage.
• Practical advantages: sequential updating enables laboratories to adopt BHM incrementally and to process tests one-at-a-time using priors derived from historical posteriors, avoiding cumbersome reanalysis of the entire archive and maintaining routine lab workflows.
• Statistical issues addressed: the Bayesian framework mitigates problems arising from very small df in frequentist residual-variance estimation (inverse-chi-square behavior), and Stein-type shrinkage improves multivariate estimation when several concentrations are estimated simultaneously.

Benefits and practical applications

• Improved accuracy and consistency of microcystin concentration estimates, reducing false positives/negatives in public-health decisions during HAB events.
• No changes required in laboratory wet-lab protocols; the method is an analytical/statistical replacement for the curve-fitting and inversion step.
• Computational workflow can be automated (e.g., a Shiny app) so non-statistical personnel can apply BHM and sequential updating readily.
• Methodology is generalizable to other calibration-based analytical methods where repeated small-sample calibrations are performed (ligand-binding assays, ELISA variants, other plate-based assays).

Future trends and potential applications

• Operational deployment: development of user-friendly software tools (Shiny applications or integrated LIMS modules) to perform BHM with sequential updating and to present uncertainty quantification to decision-makers in near-real time.
• Adaptive priors and discounting: incorporate time-varying discount factors for older historical data, preventing outdated calibration behavior from degrading current estimates; dynamic weighting strategies from Bayesian time-series methods can be applied.
• Extension to multi-analyte and multi-platform calibration: hierarchical sharing of information across assays or instruments, and cross-calibration between platforms (e.g., ELISA vs LC–MS) using shared hierarchical structure.
• Regulatory adoption: inclusion of Bayesian uncertainty metrics in guidance for water-quality advisories could improve decision robustness and reduce unnecessary public-health interventions.

Conclusions

The study demonstrates that Bayesian hierarchical modeling with sequential updating materially improves the accuracy and stability of ELISA-derived microcystin estimates compared to standard inverse-function calibration. Key practical recommendations are to use all available standard replicates (12 points per plate) and to apply shrinkage models when computationally feasible. Sequential updating enables straightforward adoption in routine laboratory environments with no changes to wet-lab procedures and modest computational cost per test. These improvements can reduce public-health risk and operational costs related to HAB events and are broadly applicable to other calibration-dependent assays.

Reference

Babica P., Bláha L., Maršálek B., 2006. Exploring the natural role of microcystins—a review of effects on photoautotrophic organisms. J. Phycol. 42, 9–20.
Bernardo J.M., Smith A.F.M., 2009. Bayesian Theory. John Wiley & Sons.
Bingham M., Sinha S.K., Lupi F., 2015. Economic Benefits of Reducing Harmful Algal Blooms in Lake Erie. Environmental Consulting and Technology, Inc.
Biosensis, 2024. Technical Note 2, Data Analysis for ELISA Assays.
Boegehold A.G., et al., 2023. Routine monitoring of western Lake Erie to track water quality changes associated with cyanobacterial harmful algal blooms. Earth Syst. Sci. Data Discuss.
Box G.E.P., Tiao G.C., 1973. Bayesian Inference in Statistical Analysis. Addison-Wesley.
Chorus I., Welker M., 2021. Toxic Cyanobacteria in Water–a Guide To their Public Health Consequences, Monitoring and Management. Taylor & Francis.
Efron B., 1975. Biased versus unbiased estimation. Adv. Math. 16, 259–277.
Efron B., 1996. Empirical Bayes methods for combining likelihoods. J. Amer. Statist. Assoc. 91, 538–550.
Findlay J.W., Dillard R.F., 2007. Appropriate calibration curve fitting in ligand binding assays. AAPS J. 9, 29.
Gan N., et al., 2012. The role of microcystins in maintaining colonies of bloom-forming Microcystis spp. Environ. Microbiol. 14, 730–742.
Gelman A., Carlin J.B., Stern H.S., Dunson D.B., Vehtari A., Rubin D.B., 2014. Bayesian Data Analysis, 3rd ed. CRC Press.
Gelman A., Chew G.L., Shnaidman M., 2004. Bayesian analysis of serial dilution assays. Biometrics 60, 407–417.
Gelman A., Jakulin A., Pittau M.G., Su Y.S., 2008. A weakly informative default prior distribution for logistic and other regression models. Ann. Appl. Stat. 2, 1360–1383.
Huisman J., et al., 2018. Cyanobacterial blooms. Nat. Rev. Microbiol. 16, 471–483.
Klauenberg K., Walzel M., Ebert B., Elster C., 2015. Informative prior distributions for ELISA analyses. Biostatistics 16, 454–464.
Miller J.N., Miller J.C., 2010. Statistics and Chemometrics for Analytical Chemistry, 6th ed. Pearson.
Nummer S.A., et al., 2018. Updating the ELISA standard curve fitting process to reduce uncertainty in estimated microcystin concentrations. MethodsX 5, 304–311.
Qian S.S., 2016. Environmental and Ecological Statistics with R. Chapman & Hall/CRC.
Qian S.S., Chaffin J.D., DuFour M.R., Sherman J.J., Golnick P.C., Collier C.D., Nummer S.A., 2015. Quantifying and reducing uncertainty in estimated microcystin concentrations from the ELISA method. Environ. Sci. Technol. 49, 14221–14229.
Qian S.S., Reckhow K.H., 2007. Combining model results and monitoring data for water quality assessment. Environ. Sci. Technol. 41, 5008–5013.
Qian S.S., Jaffe S., Gionfriddo E., Wang H., et al., 2025. Rethinking calibration as a statistical estimation problem to improve measurement accuracy. Analytica Chimica Acta (in review).
Stan Development Team, 2022. Stan modeling language user’s guide and reference manual; RStan interface.
Stein C., 1956. Inadmissibility of the usual estimator for the mean of a multivariate normal distribution. Proc. Third Berkeley Symp. Math. Statist. Prob., Univ. California Press.
Svirčev Z., et al., 2017. Toxicology of microcystins with reference to human intoxications and epidemiological investigations. Arch. Toxicol. 91, 621–650.
Toledo Department of Public Utilities, 2014. Microcystin event preliminary summary.
West M., Harrison J., 1997. Bayesian Forecasting and Dynamic Models, 2nd ed. Springer.