In recent years, the field of toxicology has witnessed a profound transformation, moving from animal-based experiments to the use of New Approach Methodologies (NAMs). These methods, which include in vitro, ex vivo and in silico systems, allow researchers to test hypotheses about chemical safety while adhering to the 3Rs principle (Replacement, Reduction and Refinement of animal use). Yet, despite their promise, NAMs come with a big challenge: integrating multiple results derived from different sources into a single, coherent conclusion. This is the main focus of the ongoing debate in Next Generation Risk Assessment (NGRA).
The integration of NAMs is not as straightforward as it might appear. Each methodology provides a piece of the puzzle, often of varying quality, relevance, and uncertainty. When these fragments are combined inconsistently, the outcome can be misleading, potentially jeopardising regulatory decisions. Until recently, such integration relied largely on expert judgement, a process valuable for its experience but limited by subjectivity and poor reproducibility. The challenge, therefore, is to ensure transparency, traceability and robustness in how evidence is combined and interpreted. This is where TOXTRUST comes into play.
TOXTRUST is an open-source computational tool designed to make sense of complex toxicological data by applying the mathematical framework of the Dempster–Shafer Theory (DST) [1, 2]. Developed at the Universitat Pompeu Fabra in Barcelona, the tool provides a transparent and structured approach to combining evidence and reasoning under uncertainty. The Theory of Belief Functions (DST) allows probabilities from different sources to be combined without assuming that one is more correct than another. Rather than forcing a single outcome, it expresses conclusions through probability bounds, showing both what is known (the “belief”) and what remains plausible but uncertain.
For decades, the DST has been recognised for its theoretical power in managing uncertainty. However, its real-world application was hindered by mathematical complexity and the need for numerous parameters and equations. TOXTRUST overcomes these barriers by offering a user-friendly interface that guides users through the process of integrating data and evaluating confidence in the results. The tool, written in Python and equipped with an intuitive graphical interface, transforms abstract theory into a practical solution for risk assessors. In simple terms, it allows toxicologists to integrate different lines of evidence, from in vitro tests, computational models or expert judgement, in a consistent and transparent way.
The strength of TOXTRUST lies not only in its mathematical soundness but also in its accessibility. Users can enter results from different NAMs, assign reliability and relevance scores, and select combination rules that best reflect the nature of the evidence. The software then computes the degrees of belief and plausibility for each outcome, clearly indicating how certain (or uncertain) a conclusion is. Instead of providing a yes-or-no answer, it delivers a probability range, which is a more honest reflection of scientific reality, where uncertainty is the rule rather than the exception.
By integrating the Dempster–Shafer Theory into a visual and interactive environment, TOXTRUST supports the transparent justification of decisions in chemical risk assessment. It shifts the focus from subjective interpretation to reproducible, mathematically grounded reasoning. This approach represents a step forward in combining evidence and reasoning with uncertainty, ensuring that risk assessment decisions are not only scientifically sound but also explainable and trustworthy.
In a landscape increasingly rich in data yet complex in interpretation, we believe that tools like TOXTRUST are essential for the future of regulatory science. They bridge the gap between advanced computational theory and practical toxicology, reinforcing the credibility and transparency of safety decisions. As the use of NAMs continues to grow, embracing structured and transparent data integration frameworks will be vital to ensure that innovation in toxicology goes hand in hand with reliability and public trust.
“Science is built up with facts, as a house is with stones; but a collection of facts is no more a science than a heap of stones is a house” – Henri Poincaré