CoChat

Optimal Experimental Design for Parameter Estimation of a Cell Signaling Model

Open access

The DOI matches its Crossref record; 34 of the 34 references it cites were found in a registry or an index; 10 of the 22 sources its claims cite were checked for support.

Abstract

Differential equation models that describe the dynamic changes of biochemical signaling states are important tools to understand cellular behavior. An essential task in building such representations is to infer the affinities, rate constants, and other parameters of a model from actual measurement data. However, intuitive measurement protocols often fail to generate data that restrict the range of possible parameter values. Here we utilized a numerical method to iteratively design optimal live-cell fluorescence microscopy experiments in order to reveal pharmacological and kinetic parameters of a phosphatidylinositol 3,4,5-trisphosphate (PIP(3)) second messenger signaling process that is deregulated in many tumors. The experimental approach included the activation of endogenous phosphoinositide 3-kinase (PI3K) by chemically induced recruitment of a regulatory peptide, reversible inhibition of PI3K using a kinase inhibitor, and monitoring of the PI3K-mediated production of PIP(3) lipids using the pleckstrin homology (PH) domain of Akt. We found that an intuitively planned and established experimental protocol did not yield data from which relevant parameters could be inferred. Starting from a set of poorly defined model parameters derived from the intuitively planned experiment, we calculated concentration-time profiles for both the inducing and the inhibitory compound that would minimize the predicted uncertainty of parameter estimates. Two cycles of optimization and experimentation were sufficient to narrowly confine the model parameters, with the mean variance of estimates dropping more than sixty-fold. Thus, optimal experimental design proved to be a powerful strategy to minimize the number of experiments needed to infer biological parameters from a cell signaling assay.

Overview

CoChat Summary

TL;DR

The paper applies optimal experimental design to a live-cell PI3K/PIP3 signaling assay and shows that an intuitively planned protocol yields poorly defined parameters, whereas numerically optimized drug concentration-time profiles sharply reduce parameter uncertainty. Two cycles of optimization and experimentation reduced the mean variance of estimates from 193.5 to 2.9, more than sixty-fold.

Key findings

  • The established protocol using iRap and LY294002 produced high parameter uncertainty and strong covariance; the initial fit had a mean variance of 193.5 and relative uncertainties indicating the intuitive protocol was not suitable for defining parameters of interest.
  • A first numerically optimized experiment was predicted to reduce the mean of variances more than 50-fold compared with repeating the established protocol; after two optimization/experimentation cycles, the mean variance dropped from 193.5 to 2.9.
  • The optimized protocols were non-intuitive but informative: the first used low-dose iRap, then higher iRap with LY29, then LY29 washout with iRap; the second used constrained drug additions and optimized time points, with the determinant criterion focusing on the LY29 inhibitor constant.
  • Data from a single optimized experiment largely eliminated correlation between estimates, and the eigenvalue spectrum of the covariance matrix was consistently reduced, though some parameter estimates still varied strongly across experiments.

Methods

  • NIH 3T3 fibroblasts were transfected with Lyn-FRB, CF-p85, and Y-PH constructs, serum-starved, and imaged by live-cell confocal fluorescence microscopy; cytosolic fluorescence of CF-p85 and Y-PH was quantified from single cells, with the traditional protocol using 25 frames captured every minute.
  • A system of ordinary differential equations modeled PIP3, CF-p85/iRap/Lyn-FRB complex formation, and Y-PH membrane binding; parameters were estimated by weighted least squares using a multiple shooting Gauss-Newton algorithm, with measurement errors estimated from individual frames.
  • Optimal experimental designs were computed by minimizing functions of the parameter covariance matrix, using sequential quadratic programming in VPLAN/PARFIT; the trace criterion was used for the first optimized design, and the determinant criterion under dynamic constraints for the second.

Limitations

  • The paper states that parameter estimates varied strongly from experiment to experiment, with most relative uncertainties remaining above 100%, and that the exogenous parameter describing CF-p85 association to Lyn-FRB increased in uncertainty because Lyn-FRB is not fluorescently tagged and is virtually unidentifiable.
  • The paper notes that minimizing the determinant can focus experimental effort on a single parameter and be sensitive to overall uncertainty, and that experimental imperfection such as residual LY29 after washout introduced uncertainty, though it did not significantly affect the realized design performance.
  • The paper states that determined parameter values are model-specific and that future optimization methods need to address robustness to parameter uncertainty, model structure, and experimental uncertainty; evident from the text, the model also relies on simplifying assumptions such as irreversible binding steps and Y-PH not shielding PIP3 from degradation.

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

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1000558

Cites

34 of 34 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Figures

Figure 1 Parameter estimation of PI3K signaling reveals identifiability problems for important kinetic properties. (A) iRap-induced dimerization of FKBP with FRB recruits CF-p85 to the plasma membrane and activates the catalytic subunit p110 of PI3K to produce from . PH domains of many downstream signaling proteins bind to enriched membranes with differential affinities. LY29 is a widely used inhibitor of PI3K. (B) Confocal fluorescence images of an NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). Initially, CF-p85 and Y-PH reside in the cytosol (left). Addition of iRap recruits both constructs to the plasma membrane as can be judged from cytosolic depletion (middle). Y-PH but not CF-p85 dissociates from the plasma membrane upon addition of LY29. Experiment time is shown in minutes. (C) Differential equations model of the dynamic system illustrated in (A). System parameters are shown in blue, variables that allow control of the system are shown in green, and calibration parameters which were different from cell to cell are shown in gold. (D) Parameter values can be inferred by minimizing the discrepancy between model and experimental data. The schematic illustrates this concept for two parameters, and . The range of possible parameter values is where this discrepancy falls below the confidence level (blue shaded ellipsoid). If this range is large, parameters can be inferred only with high uncertainty. (E) Fluorescence was quantified in the green square depicted in (B), and the model in (C) was fit to that data. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange error bars and lines cytosolic fluorescence of Y-PH. (F) The relative uncertainty of parameter estimates was very high. This is a common problem in systems biology models. (G) Schematic projection of the confidence ellipsoid from (D). Good experiments minimize the size of this uncertainty region (shaded blue). Criteria that have been proposed for describing this size are the longest half-axis, the average extent, or the volume of the confidence ellipsoid, among others. These criteria correspond to properties of the parameter covariance matrix .
Figure 1 Parameter estimation of PI3K signaling reveals identifiability problems for important kinetic properties. (A) iRap-induced dimerization of FKBP with FRB recruits CF-p85 to the plasma membrane and activates the catalytic subunit p110 of PI3K to produce from . PH domains of many downstream signaling proteins bind to enriched membranes with differential affinities. LY29 is a widely used inhibitor of PI3K. (B) Confocal fluorescence images of an NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). Initially, CF-p85 and Y-PH reside in the cytosol (left). Addition of iRap recruits both constructs to the plasma membrane as can be judged from cytosolic depletion (middle). Y-PH but not CF-p85 dissociates from the plasma membrane upon addition of LY29. Experiment time is shown in minutes. (C) Differential equations model of the dynamic system illustrated in (A). System parameters are shown in blue, variables that allow control of the system are shown in green, and calibration parameters which were different from cell to cell are shown in gold. (D) Parameter values can be inferred by minimizing the discrepancy between model and experimental data. The schematic illustrates this concept for two parameters, and . The range of possible parameter values is where this discrepancy falls below the confidence level (blue shaded ellipsoid). If this range is large, parameters can be inferred only with high uncertainty. (E) Fluorescence was quantified in the green square depicted in (B), and the model in (C) was fit to that data. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange error bars and lines cytosolic fluorescence of Y-PH. (F) The relative uncertainty of parameter estimates was very high. This is a common problem in systems biology models. (G) Schematic projection of the confidence ellipsoid from (D). Good experiments minimize the size of this uncertainty region (shaded blue). Criteria that have been proposed for describing this size are the longest half-axis, the average extent, or the volume of the confidence ellipsoid, among others. These criteria correspond to properties of the parameter covariance matrix .CC BY · PubMed Central
Figure 2 Numerical optimization of experimental design and practical experimentation to infer model parameters from live-cell microscopy data. (A) Schematic of sequential experimentation. The traditional design (left) resulted in a model that was used for optimizing the concentration-time profile of iRap (green) and LY29 (pink) for the next experiment. (B) The washout of LY29 at minute 15 was realized by inserting a needle connected to a vacuum trap, and rinsing with 5 ml of iRap in extracellular buffer. (C) Mean of parameter variances predicted for or achieved with the optimized experiment. Arrows indicate the parameter variance at different stages of model development, and under consideration of experimental uncertainty introduced by the wash. (D) Confocal fluorescence images of an NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). 30 nM iRap was added immediately before triggering time-lapse acquisition. After 5 min, iRap and LY29 were added. 10 min later, LY29 was washed out, using iRap in extracellular buffer. Experiment time is shown in minutes. (E) Combined parameter estimation from both the traditional (left) and optimized (right) experiment. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange error bars and lines cytosolic fluorescence of Y-PH. The frame captured at minute 15 of the optimized experiment was lost to brightfield illumination during the washout procedure.
Figure 2 Numerical optimization of experimental design and practical experimentation to infer model parameters from live-cell microscopy data. (A) Schematic of sequential experimentation. The traditional design (left) resulted in a model that was used for optimizing the concentration-time profile of iRap (green) and LY29 (pink) for the next experiment. (B) The washout of LY29 at minute 15 was realized by inserting a needle connected to a vacuum trap, and rinsing with 5 ml of iRap in extracellular buffer. (C) Mean of parameter variances predicted for or achieved with the optimized experiment. Arrows indicate the parameter variance at different stages of model development, and under consideration of experimental uncertainty introduced by the wash. (D) Confocal fluorescence images of an NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). 30 nM iRap was added immediately before triggering time-lapse acquisition. After 5 min, iRap and LY29 were added. 10 min later, LY29 was washed out, using iRap in extracellular buffer. Experiment time is shown in minutes. (E) Combined parameter estimation from both the traditional (left) and optimized (right) experiment. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange error bars and lines cytosolic fluorescence of Y-PH. The frame captured at minute 15 of the optimized experiment was lost to brightfield illumination during the washout procedure.CC BY · PubMed Central
Figure 3 Minimization of different uncertainty criteria under experimentally meaningful design constraints. Different uncertainty criteria were minimized by finding optimal time points for adding drugs and optimal concentrations. Washouts were excluded from the design space by a dynamic constraint formulation. (A) Performance comparison of different criteria with respect to parameter uncertainty or (B) the eigenvalue spectrum of the optimized experimental design. Arrows indicate the eigenvalue that is aligned with the uncertainty of . (C) Schematic of sequential experimentation. The data from the traditional and the first optimized experiment were used for optimizing the concentration-time profile of iRap (green) and LY29 (pink) for the next experiment. (D) NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). iRap was added at the beginning of the experiment. 10 min later, another 10 nM LY29 and another iRap were added. PI3K remained active as judged from low cytosolic fluorescence of Y-PH. LY29 was added at minute 20, resulting in dissociation of Y-PH from the plasma membrane. Time is shown in minutes. (E) Combined parameter estimation on all three experiments. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange ones show cytosolic fluorescence of Y-PH.
Figure 3 Minimization of different uncertainty criteria under experimentally meaningful design constraints. Different uncertainty criteria were minimized by finding optimal time points for adding drugs and optimal concentrations. Washouts were excluded from the design space by a dynamic constraint formulation. (A) Performance comparison of different criteria with respect to parameter uncertainty or (B) the eigenvalue spectrum of the optimized experimental design. Arrows indicate the eigenvalue that is aligned with the uncertainty of . (C) Schematic of sequential experimentation. The data from the traditional and the first optimized experiment were used for optimizing the concentration-time profile of iRap (green) and LY29 (pink) for the next experiment. (D) NIH 3T3 cell (CF-p85 shown in cyan, Y-PH in yellow). iRap was added at the beginning of the experiment. 10 min later, another 10 nM LY29 and another iRap were added. PI3K remained active as judged from low cytosolic fluorescence of Y-PH. LY29 was added at minute 20, resulting in dissociation of Y-PH from the plasma membrane. Time is shown in minutes. (E) Combined parameter estimation on all three experiments. Error bars show data, continuous lines show model output. Blue error bars and lines show cytosolic fluorescence of CF-p85, orange ones show cytosolic fluorescence of Y-PH.CC BY · PubMed Central
Figure 4 Decline of uncertainty. (A) The mean of parameter variances dropped from 193.5 after the traditional experiment to 2.9 after the second optimized one. For comparison, parameter uncertainties were calculated for repetitions of the intuitive traditional experiment. (B) Parameter uncertainties after individual experiments were performed. (C) Normalized eigenvectors of the corresponding covariance matrices ordered in ascending length suggest that correlation between biological parameters was largely eliminated. (D) The ranges of the eigenvalue spectra of were consistently reduced.
Figure 4 Decline of uncertainty. (A) The mean of parameter variances dropped from 193.5 after the traditional experiment to 2.9 after the second optimized one. For comparison, parameter uncertainties were calculated for repetitions of the intuitive traditional experiment. (B) Parameter uncertainties after individual experiments were performed. (C) Normalized eigenvectors of the corresponding covariance matrices ordered in ascending length suggest that correlation between biological parameters was largely eliminated. (D) The ranges of the eigenvalue spectra of were consistently reduced.CC BY · PubMed Central

Verification

Claims the paper makes

6 Valid · 2 Not valid · 11 Not checked · 74 Own results and interpretations

43 more claims

Create a free account to see all 93 claims and to verify the 12 references still open.

See all claims — free account
  • Probably not a substantial claim. 24
  • It may state two separate claims as one. 22
  • The quote only partly states this claim. 2

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?

Citations

Cites

Show all 34 references

Cited by

See all 135

Metadata

Authors
Bandara, S., Schlöder, J. P., Eils, R., Bock, H. G., & Meyer, T.
Affiliations
  • Stanford University (US)
  • Heidelberg University (DE)
  • German Cancer Research Center (DE)
Corresponding author
  • Bandara, Samuel
  • Meyer, Tobias
Venue
PLoS Computational Biology
Published
2009-11-06
Volume
5
Issue
11
Pages
e1000558
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W1979892077
Cited by
135
References
34
Citations by year
2026: 1 · 2025: 5 · 2024: 5 · 2023: 6 · 2022: 5 · 2021: 5 · 2020: 5 · 2019: 7 · 2018: 12 · 2017: 9 · 2016: 8 · 2015: 9 · 2014: 10 · 2013: 14 · 2012: 14
FWCI
4.57
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
  • Microsoft Research
  • National Institutes of Health (MH064801)
  • National Institute of Mental Health (R01MH064801, MH 064801)
  • National Institute of General Medical Sciences (R01GM048113, R01GM063702, R01GM030179, GM063702)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Social Sciences › Decision Sciences › Management Science and Operations Research › Optimal Experimental Design Methods
  • Physical Sciences › Engineering › Control and Systems Engineering › Control Systems and Identification
Keywords
optimal experimental design, parameter estimation, signal transduction, PIP3 signaling, phosphatidylinositol 3-kinase, Akt, pleckstrin homology domain, live-cell fluorescence microscopy, differential equation model, kinase inhibitors, parameter uncertainty
MeSH terms
Algorithms, Animals, Cell Communication, Kinetics, Research Design, Protein Structure, Tertiary, Microscopy, Confocal, Phosphatidylinositol 3-Kinases, NIH 3T3 Cells, Systems Biology, Mice

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

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 (qwen3.8-omni-flash and 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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Optimal Experimental Design for Parameter Estimation of a Cell Signaling Model. https://app.cochat.ai/papers/10.1371/journal.pcbi.1000558
The paper
Bandara, S., Schlöder, J. P., Eils, R., Bock, H. G., & Meyer, T. (2009). Optimal Experimental Design for Parameter Estimation of a Cell Signaling Model. PLoS Computational Biology. https://doi.org/10.1371/journal.pcbi.1000558

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.