Expanding the Toolbox: Hazard-Screening Methods and Tools for Identifying Safer Chemicals in Green Product Design

A key focus of green product design is to reduce the product's inherent chemical hazard. Various alternative assessment methodologies may be used to compare the hazard properties of possible candidate chemicals. However, only a small fraction of the chemicals currently in commercial use are adequately characterized in terms of toxicological effects. This limitation can hamper the study of safer chemical alternatives and increase the likelihood of regrettable substitutions. Approaches for addressing such data gaps include read-across, in silico programs and high throughput in chemico and in silico assays. Each of these show considerable promise although a consensus on how to use them for hazard evaluation of data poor chemicals is lacking. The limitations of such tools, which attempt to simplify complex biology into key predictive factors, is also often underestimated. To evaluate currently available approaches for addressing data gaps, we established three test sets of chemicals, each with structural similarity to a target chemical (target chemical 1: 4-phenylenediamine, target chemical 2: hydroxyethyl acrylate, target chemical 3: methylisothiazolone). We first compared results from the in silico programs Toxtree and Derek Nexus with animal test data obtained using standard assays. We then compared chemical similarity scores calculated by two computational tools Toxmatch and ChemMine. Lastly, we refined our test sets by applying a series of exclusion criteria, including in silico analysis and physicochemical data relevant for skin sensitization (e.g., molecular weight, water solubility, and vapor pressure). The in silico programs in combination exhibited a sensitivity of 92% and specificity of 88%. Toxmatch and ChemMine demonstrated good agreement in their similarity score rankings across the three test sets (TS1: W = 0.74, p = 0.014; TS2: W = 0.72, p = 0.067; TS3: W = 0.87, p = 0.095). Narrowing our chemical test sets using physical chemical properties and in silico evaluation improved the overall accuracy of our read-across approach compared with the initial unrefined test sets (TS1: 56% improved to 100%; TS2: 54% to 100%; TS3: 50% to 100%). Our findings support the development of robust read-across approaches incorporating available data-gap filling tools to help conduct screening level alternatives assessments and identify safer chemicals as part of green product design.