Capturing\nDifferential Quality of Experimental Evidence\nin a Predictive Quantum-Mechanical Model for Respiratory Sensitization

Asthma is of concern in occupational toxicology with\nsignificant\npublic-health and economic costs. In the absence of benchmark in vivo\nand in vitro tests, the use of mechanistically sound in silico models\nis critical to inform hazard and to protect workers from exposure\nto potentially harmful substances. We recently reported on the computer-aided\ndiscovery and REdesign (CADRE) model for respiratory sensitization,\nwhich relies on a tiered structure of expert rules, molecular simulations,\nquantum-mechanics calculations and advanced statistics to accurately\nidentify respiratory sensitizers from first principles. Here, we present\nan update to this model based on two years of testing in the pharmaceutical\nspace, where we captured the heterogeneity of the underlying experimental\nevidence in two predictive tiers, thus allowing the practitioner to\nselect an outcome based on their expert assessment of the data reliability\nand relevance. This user-based tuning of predictive models is critical\nfor end points that lack consensus on what constitutes satisfactory\nevidence to support a decision in the handling of chemicals for occupational\nsafety.