- Data-Driven Computational Approaches to Complex Chemical Mixtures
-
Hemisfair Ballroom C3‚ Convention Center
Chair: Adrian Green, Sciome LLC
Co-Chair: David Reif, NIEHS/DTT
Primary Endorser: Mixtures Specialty Section
Endorser(s): Computational Toxicology Specialty SectionUnderstanding and predicting the toxicity of complex chemical mixtures remains one of the most persistent challenges in modern toxicology and risk assessment. Traditional approaches often struggle to keep pace with the combinatorial explosion of possible interactions, particularly when mixtures contain dozens of constituents or fall into poorly defined categories such as UVCBs (Unknown or Variable composition, complex reaction products, or natural products). In this session, we bring together emerging conceptual frameworks and computational strategies that aim to transform how scientists evaluate mixture toxicity, moving from qualitative, expert-driven assessments toward scalable, quantitative, and reproducible methodologies.
We begin by exploring a novel perspective on mixture toxicity that proposes to determine similarity between two mixtures by using a combination of chemical-specific toxicity and a chemical’s concentration in a mixture to weigh the importance of differences in the chemical’s concentrations between two mixtures. The approach also identifies the unique binary pairs of a mixture and ranks the pairs in terms of the likelihood that they will interact and will contribute meaningfully to an adverse outcome. This strategy has the potential to reduce the analytical burden and prioritize the contributions of biologically relevant drivers of mixture toxicity.
Building on this foundation, we examine how artificial intelligence and machine learning can be leveraged to assess constituent contributions to toxicity. These approaches enable the systematic integration of large, multidimensional datasets, uncovering patterns and relationships that are difficult to detect using conventional methods. By quantifying the relative importance of individual components and their interactions, AI/ML models offer powerful tools for predicting mixture effects, supporting risk assessment, and guiding experimental design.
Finally, we address the critical challenge of similarity assessment for complex substances such as petroleum-derived UVCBs. Using hydrocarbon block (HCB) representations derived from advanced chromatographic data, we demonstrate how both composition-based and descriptor-based methods can quantify similarity across hundreds of substances. These complementary approaches, one grounded in compositional profiles and the other in molecular descriptors, provide nuanced insights into chemical relationships, supporting read-across applications and regulatory decision-making.
Together, these presentations highlight a shift toward more structured, data-driven approaches for understanding complex mixtures. Attendees will learn how to combine prioritization strategies, machine learning, and quantitative similarity metrics to tackle one of toxicology’s most difficult problems. Whether working with defined mixtures or highly variable substances, this session will provide practical frameworks and forward-looking strategies to advance mixture risk assessment in both research and regulatory contexts.Session and Activity Type
Enter Note
Go to previous page in this tab
Event
