Mixture Touch: A Web Platform for the Evaluation of Complex Chemical Mixtures

Risk assessment of chemical mixtures isRisk assessment of chemical mixtures is challenging because information about the chemical structure, concentration, properties, and toxicity, down to the individual compounds, is generally not readily accessible. To cope with this challenge, we think Mixture Touch― a web platform that offers a one-window solution, for free, for the risk assessment of complex mixtures that are analyzed with comprehensive two-dimensional gas chromatography (GC × GC). GC × GC is a powerful analytical technique for target and nontarget analysis of complex mixtures. Our web platform allows users to visualize the GC × GC data, conduct spectral identification, estimate properties, and analyze potential risks based on established methods. For illustration purpose, we show how to assess the aquatic bioaccumulation potential of short-chain chlorinated paraffin (SCCP), which is an industrially manufactured mixture. The platform readily demonstrated that most of the SCCP congeners did not have the tendency to accumulate in aquatic organisms but in humans. The platform can bridge the gap between the GC × GC experts, GC × GC users, analytical experts, and risk assessors. It could enhance the level of risk assessments of mixtures utilizing the high performance of the state-of-the-art analytical instruments.


Section S-1. Mechanism of GC × GC property estimation
The partitioning behavior of a compound between two different phases is expressed by the Gibbs free energy (ΔG) of phase transfer for the target compound. The logarithm of the equilibrium constant between the phases (log K xy ) is linearly correlated with the energy of phase transfer, known as the linear free energy relationship (LFER) shown in Eq. (1).
where R is the ideal gas constant, and T is temperature of the system.
In the polyparameter LFER (pp-LFER) model, the log K xy is expressed by a linear combination of descriptors (E, S, A, B, L) of the chemical with system constants (e, s, a, b, l), as shown in Eq. (2).
where E is the descriptor for excess molar refraction, S is the descriptor for the molecular polarizability and polarity, A is the hydrogen-bond acidity, B is the hydrogen-bond basicity, and L is the logarithmic partition coefficient between air and hexadecane. There are several forms of the pp-LFER model, which are described in detail elsewhere (Endo & Goss, 2014;Panagopoulos et al., 2015).
Nabi et al. developed a two-parameter LFER, which uses descriptors extracted from the GC × GC retention times (Nabi & Arey, 2017;Nabi et al., 2014). For nonpolar chemicals, fewer dimensions are required to model log K xy . Descriptors of the original pp-LFER were consolidated based on a singular value decomposition to two retention time descriptors (u 1 and u 2 ), which were linked to partitioning coefficients in the first and second GC columns (Arey et al., 2005;Nabi et al., 2014). Eventually, properties related to partitioning can be expressed by the position of each non-polar chemical in the GC × GC as the following equations: log K xy,i = λ 1 u 1,i + λ 2 u 2,i + λ 3 …(3) where  1-3 are the calibrated parameters from a multiple linear regression analysis in this study. The parameter β orth was determined by Gram-Schmidt orthogonalization. Parameters α 1 and α 2 were calibrated with the ideal carbon number ( ) for selected non-polar chemicals using linear regression analysis. *

The
for each chemical was calculated by linear interpolation of the retention * times of alkanes (C 10 -C 50 ) in the first dimension. Parameters , , and α 3 were ′ 1 ′ 2 calibrated using , , and and non-linear regression analysis. The * 2, * 2, parameter is the retention time of the selected chemical in the second column,

2,
and is the RT2 of the ideal carbon where RT1 corresponds to that of a target * 2, compound. Further details for these calculations can be found elsewhere (Nabi & Arey, 2017;Nabi et al., 2014;Zushi et al., 2019). The Mixture Touch is equipped with an equation for 12 environmentally-relevant properties including logarithms of the partition coefficients of octanol-water (logK o-w ) and air-water (log K a-w ). The equation was calibrated with non-polar compounds, which were defined as structures containing carbon, hydrogen, and halogen atoms.

Section S-2. Algorithm of spectrum search
The Mixture Touch is equipped with a mass spectrum search function. A reference spectrum that shows a similar pattern to the target spectrum is picked from a database according to the degree of similarity. Practically, the similarity is calculated using the following equation: where Z x is the normalized intensity of the m/z of the target spectrum, and Z y is its counterpart spectrum. Z is the matrix of Z x and Z y . The mass spectrum is generally evaluated using the NIST MS search program. According to the guidelines built into the NIST software, a value of 900 or greater is an excellent match, 800-900 is a good match, 700-800 is a fair match, and less than 600 is a very poor match. The above equation corresponds to the setting of the NIST software of 'simple similarity' with 'default' pre-search.
The URLs for external websites that also provide spectrum search functions with their own spectral libraries are available in the "Spectrum Search" tab. In its