A classification-based methodology for the estimation of binary surfactant critical micelle concentrations
Chetan R. Chilkunda, John R. Kitchin, and Robert D. Tilton, “A classification-based methodology for the estimation of binary surfactant critical micelle concentrations”, Digital Discovery 4(6), 1449-1456 (2025). https://doi.org/10.1039/d5dd00058k
This summary was written by a language model from the paper and has not been checked line by line. Rely on the published paper for details.
Note on the source text: the text available to this wiki is the accepted manuscript version. It reads cleanly apart from two-column interleaving; figures are not available, so results are described from the prose and the table.
Question
Commercial products such as detergents combine surfactants (for example anionic ones for solubilizing power and nonionic ones for hardness tolerance), and their mixtures are usually non-ideal. Characterizing the critical micelle concentration (CMC) of a mixture across all compositions conventionally means finding a CMC at several fixed ratios and then fitting a model through those points, which takes many experiments. The paper asks whether a single 96-well plate, planned with a physical model and analyzed as a classification problem, can determine the mixture CMC boundary and the surfactant interaction parameter (Chilkunda et al. 2025).
Approach
Regular solution theory serves as the physical model. It relates the mixed CMC to the pure-component CMCs, the bulk and micellar mole fractions, and a single empirical interaction parameter beta, with beta = 0 for ideal mixing and negative values for attractive (synergistic) interactions. The test system was anionic sodium dodecyl sulfate (SDS) with nonionic tetraethylene glycol monooctyl ether (C8E4), chosen because a literature value of beta = -3.1 exists. Micellization was detected by dye solubilization: Nile Red fluoresces weakly in water and strongly once it partitions into micelle cores, and readings were normalized to a 10 mM SDS internal standard on a plate spectrofluorometer (Chilkunda et al. 2025).
The design of experiments first fixed the pure CMCs, then drew five constant-ratio lines through composition space and placed trials where each line crosses the regular solution curves for seven beta values in the restricted range from -6 to 0 (expected for anionic plus oligo(ethylene oxide) nonionic pairs). Two extra trials below the lowest curve (at 10% and 20% of the ideal-mixing point) and one above the ideal line bracket each line. With duplicates for all but the top trial this gives 95 wells plus the standard, filling one plate (Chilkunda et al. 2025).
Two analyses were compared. The first fits a leaky ReLU to fluorescence along each line to get five CMCs, then fits regular solution theory to those plus the two pure CMCs by orthogonal distance regression. The second labels every well as below (0) or above (1) the CMC using a plane fitted to the low-concentration trials and extrapolated, with its 95% confidence interval, as a fluorescence threshold; it then finds the range of beta whose regular solution boundary minimizes misclassifications across all data at once (Chilkunda et al. 2025).
Key findings
Measured pure CMCs were 8.0 +/- 0.2 mM for SDS and 9.3 +/- 0.3 mM for C8E4, within 2.5% of literature values, and in 0.5 M NaCl they were 0.6 +/- 0.1 mM for SDS (literature range 0.5 to 0.9 mM) and 7.0 +/- 0.2 mM for C8E4 (Chilkunda et al. 2025). A clear fluorescence jump near a normalized intensity of about 0.2 marked the onset of micellization across the plate. The two-step leaky ReLU analysis gave beta = -2.8 +/- 0.2, but each line is fit separately, it is sensitive to local noise and initial guesses, and the final regression uses only seven points. The classification analysis gave beta = -3.6 +/- 0.5 (interval -4.1 to -3.1), a 14.9% difference from the literature -3.1. In 0.5 M NaCl, where the SDS sampling range is an order of magnitude narrower than for C8E4, classification gave beta between -3.45 and -2.71 (-3.1 +/- 0.4), 17.5% from the -2.6 reported for the related SDS-C12E8 system in the same salt. The authors attribute the less negative beta to electrostatic screening of the sulfate headgroups by salt, which leads to less ethylene oxide incorporation and interaction among headgroups in the micelles, while noting that the two beta intervals overlap at 95% confidence (Chilkunda et al. 2025).
Significance
The work recasts parameter estimation for a nonlinear phase boundary as a classification problem: the physical model defines the decision boundary, and the boundary is fit to labeled observations rather than to separately extracted CMC points. The authors argue this is less biased and more robust to noise than per-line fits, that any physical model or empirical expression could replace regular solution theory, and that a single plate makes it a practical screening tool for binary surfactant mixtures. They identify ternary mixtures and other electrolyte and solvent effects as next steps. Data and analysis notebooks are on Figshare (Chilkunda et al. 2025).