CoChat

Parameter Estimation and Model Selection in Computational Biology

Open access
Lillacci, Gabriele, Khammash, Mustafa · 2010 · PLoS Computational Biology

The DOI matches its Crossref record; 45 of the 48 references it cites were found in a registry or an index; 1 of the 13 sources its claims cite were checked for support.

Abstract

A central challenge in computational modeling of biological systems is the determination of the model parameters. Typically, only a fraction of the parameters (such as kinetic rate constants) are experimentally measured, while the rest are often fitted. The fitting process is usually based on experimental time course measurements of observables, which are used to assign parameter values that minimize some measure of the error between these measurements and the corresponding model prediction. The measurements, which can come from immunoblotting assays, fluorescent markers, etc., tend to be very noisy and taken at a limited number of time points. In this work we present a new approach to the problem of parameter selection of biological models. We show how one can use a dynamic recursive estimator, known as extended Kalman filter, to arrive at estimates of the model parameters. The proposed method follows. First, we use a variation of the Kalman filter that is particularly well suited to biological applications to obtain a first guess for the unknown parameters. Secondly, we employ an a posteriori identifiability test to check the reliability of the estimates. Finally, we solve an optimization problem to refine the first guess in case it should not be accurate enough. The final estimates are guaranteed to be statistically consistent with the measurements. Furthermore, we show how the same tools can be used to discriminate among alternate models of the same biological process. We demonstrate these ideas by applying our methods to two examples, namely a model of the heat shock response in E. coli, and a model of a synthetic gene regulation system. The methods presented are quite general and may be applied to a wide class of biological systems where noisy measurements are used for parameter estimation or model selection.

Overview

CoChat Summary

TL;DR

The paper proposes a three-stage method for parameter estimation and model selection in biological ODE models: constrained hybrid extended Kalman filtering, an a posteriori variance/identifiability test, and moment-matching refinement. It demonstrates the approach on an E. coli heat shock model and a generalized repressilator model, where moment matching passed the statistical test while least-squares and genetic-algorithm fits failed.

Key findings

  • In the small-parameter heat shock case, the HEKF estimated two parameters whose true value was 3 for both; the text reports the estimation as very accurate despite high noise, and the a posteriori identifiability test at confidence coefficient 0.95 passed for both measurement-noise components.
  • For the 6-parameter heat shock case, a single HEKF run failed the identifiability test; moment-matching refinement with BFGS found a minimum after 1560 iterations with cost 0.14 in about 3 hours on a MacBook Pro with a single 2.6 GHz processor, and the refined estimates passed the test.
  • For the 18-parameter repressilator model, a single HEKF run failed the test; moment-matching refinement with BFGS took 720 iterations (about 26 minutes) and passed all three components, whereas Levenberg-Marquardt least-squares (129 iterations, about 3 minutes) and a genetic algorithm (101 iterations, about 4 minutes) failed.
  • For model selection in the heat shock example, only model \(M_1\) produced variance interval estimates containing the real measurement-noise variances; model \(M_2\), which lacks the spike in the sigma factor, was rejected as inconsistent with the measurements.

Methods

  • The model is written as a continuous-time ODE state-space system; parameters are treated as constant states via state extension, and a constrained hybrid extended Kalman filter (HEKF) handles continuous-time dynamics with discrete-time measurements and equality/inequality constraints solved as quadratic programming.
  • An a posteriori identifiability test estimates measurement-noise variances from residuals, forms point and interval estimates using the chi-square distribution, and rejects estimates whose intervals do not contain the known real noise variances; if the test fails, a moment-matching optimization minimizes mean and variance mismatch, typically with BFGS.
  • Model selection runs the HEKF on candidate models, computes measurement-noise variance estimates for each, discards models whose interval estimates exclude the real variances, and selects the model whose variances match best; examples use simulated noisy data from an E. coli heat shock model (2 and 6 parameters; 22 data points in the small case) and a generalized repressilator model (18 parameters; 30 equally spaced mRNA data points).

Limitations

  • The paper states that the extended Kalman filter is not generally optimal for nonlinear systems, may diverge or give biased estimates, can suffer from loss of observability/non-uniqueness after state extension, and is subject to numerical ill-conditioning in covariance propagation; initial estimates too far off can also cause problems.
  • The paper states that the process-noise covariance is chosen by the user, the measurement-noise statistics are assumed known, the moment-matching weights are user-selected, and the refinement optimization can be computationally expensive (e.g., about 3 hours for the 6-parameter heat shock case).
  • Evident from the text: the demonstrations use in silico simulated measurements rather than real experimental data, and many numerical entries in the tables and some reported percentages are missing from the provided text, limiting independent verification of the exact estimates.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1000696

Cites

45 of 45 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Flowchart of the proposed method. The algorithm is a three-stage process, which involves Kalman filtering, a statistical accuracy test and an optimization problem.
Figure 1 Flowchart of the proposed method. The algorithm is a three-stage process, which involves Kalman filtering, a statistical accuracy test and an optimization problem.CC BY · PubMed Central
Figure 2 Estimation of 2 parameters in the heat shock model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the HEKF estimates for the parameters and . Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graphs are zoomed to highlight the transient response of the heat shock system after a temperature increase.
Figure 2 Estimation of 2 parameters in the heat shock model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the HEKF estimates for the parameters and . Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graphs are zoomed to highlight the transient response of the heat shock system after a temperature increase.CC BY · PubMed Central
Figure 3 Time evolution of the Kalman filter parameter estimates in the heat shock model. After an initial transient, the estimates of the two parameters (top) and (bottom), represented by the triangles, keep oscillating around the respective true values (blue dashed line). The last 10 samples (connected by the green line) are averaged to extract a single number from this time-varying signal.
Figure 3 Time evolution of the Kalman filter parameter estimates in the heat shock model. After an initial transient, the estimates of the two parameters (top) and (bottom), represented by the triangles, keep oscillating around the respective true values (blue dashed line). The last 10 samples (connected by the green line) are averaged to extract a single number from this time-varying signal.CC BY · PubMed Central
Figure 4 Discrimination between competing heat shock models. The models (18) (blue) and (19) (red) are compared in terms of their (top) and (bottom) outputs. Both signals evolve to the same steady state, but with different transient behavior. The dashed lines represent the ideal model solutions, the triangles are the corresponding Kalman filter estimates.
Figure 4 Discrimination between competing heat shock models. The models (18) (blue) and (19) (red) are compared in terms of their (top) and (bottom) outputs. Both signals evolve to the same steady state, but with different transient behavior. The dashed lines represent the ideal model solutions, the triangles are the corresponding Kalman filter estimates.CC BY · PubMed Central
Figure 5 interval estimates in the case of valid and invalid parameter sets. The red set of interval estimates corresponds to a parameter set computed with the HEKF only (invalid). The green set corresponds to a parameter set that was obtained with the combination of HEKF and moment matching optimization (valid). The real variances (blue triangles) only lie inside the intervals corresponding to a valid estimation. The top panel is relative to the measurement signal, the bottom panel to the measurement signal.
Figure 5 interval estimates in the case of valid and invalid parameter sets. The red set of interval estimates corresponds to a parameter set computed with the HEKF only (invalid). The green set corresponds to a parameter set that was obtained with the combination of HEKF and moment matching optimization (valid). The real variances (blue triangles) only lie inside the intervals corresponding to a valid estimation. The top panel is relative to the measurement signal, the bottom panel to the measurement signal.CC BY · PubMed Central
Figure 6 Estimation of 6 parameters in the heat shock model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the parameters estimates. Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graphs are zoomed to highlight the transient response of the heat shock system after a temperature increase.
Figure 6 Estimation of 6 parameters in the heat shock model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the parameters estimates. Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graphs are zoomed to highlight the transient response of the heat shock system after a temperature increase.CC BY · PubMed Central
Figure 7 Estimation of 18 parameters in the repressilator model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the estimated parameters. Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graph for the measurement is presented in the supporting Figure S1.
Figure 7 Estimation of 18 parameters in the repressilator model. The data points (green squares) are obtained by evaluating the true model solution (red dashed curve) at the chosen time points, and then adding white Gaussian noise. The blue solid line shows the reconstructed solution corresponding to the estimated parameters. Both the reconstructed measurement signal for (top) and the one for (bottom) are very close to the respective true solutions. The graph for the measurement is presented in the supporting Figure S1.CC BY · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 11 Not checked · 41 Own results and interpretations

25 more claims

Create a free account to see all 54 claims and to verify the 12 references still open.

See all claims — free account
  • Probably not a substantial claim. 10
  • It may state two separate claims as one. 14
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Cited by

See all 325

Metadata

Authors
Lillacci, G., & Khammash, M.
Affiliations
University of California, Santa Barbara (US)
Corresponding author
Khammash, Mustafa
Venue
PLoS Computational Biology
Published
2010-03-05
Volume
6
Issue
3
Pages
e1000696
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W2044278781
Cited by
325
References
48
Citations by year
2026: 4 · 2025: 13 · 2024: 18 · 2023: 11 · 2022: 16 · 2021: 24 · 2020: 21 · 2019: 29 · 2018: 27 · 2017: 21 · 2016: 19 · 2015: 28 · 2014: 29 · 2013: 34 · 2012: 14
FWCI
8.46
Type
journal article
Language
en
Open access
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
Funders
  • National Science Foundation (0802008, DAAD19-03-D-0004, 0835847)
  • Deutscher Akademischer Austauschdienst
  • National Institutes of Health
  • National Institute of General Medical Sciences (R01GM049831)
  • Division of Electrical, Communications and Cyber Systems (0835847, ECCS-0802008)
  • Army Research Office (DAAD 19-03-D-0004, ECCS-0835847)
  • Institute for Collaborative Biotechnologies (ECCS-0835847, DAAD19-03-D-0004)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Physical Sciences › Engineering › Control and Systems Engineering › Control Systems and Identification
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Microbial Metabolic Engineering and Bioproduction
Keywords
parameter estimation, model selection, extended Kalman filter, recursive parameter estimation, identifiability, heat shock response, Escherichia coli, synthetic gene regulation, rate constants, mathematical biology
MeSH terms
Algorithms, Animals, Computer Simulation, Gene Expression Regulation, Humans, Models, Biological, Signal Transduction, Proteome, Systems Biology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.