SinS - Abstract

Abstract Information :

Elucidating the information contained in different molecular fingerprints Melanie Messih a, Viktoriia Turkina a, Saer Samanipour a,b,c a Van ’t Hoff Institute for Molecular Sciences (HIMS), University of Amsterdam, Amsterdam, the Netherlands b Queensland Alliance for Environmental Health Sciences (QAEHS), The University of Queensland, Australia c UvA Data Science Center The number of environmentally relevant chemicals is estimated to fall between 350,000 to 800,000, with limited knowledge of their environmental fate and toxicity. This highlights the pressing need for efficient toxicity evaluation methods that can efficiently and accurately identify the potential harm of these chemicals. Experimental toxicity assessments of such a vast number of chemicals would be unfeasible due to the extensive time, effort, and resources required. In light of this, modelling approaches have emerged as a widely accepted and applied alternative for predicting hazard indicators. Traditional modelling strategies rely on molecular descriptors, which are mathematical representations of a molecule's properties. However, the reliability of these descriptors may be compromised by their instability, resulting in potentially unreliable outcomes. As such, alternative modelling techniques such as molecular fingerprinting are being explored to improve the accuracy of toxicity predictions while reducing the need for extensive experimental assessments. Molecular fingerprints are a type of molecular descriptors that rely on the structure of the molecule, rather than its properties, and can therefore be directly associated with toxicity. There are several types of fingerprints available that describe molecular substructures in different ways. Although the information contained in different molecular fingerprints can differ considerably, little research has been conducted on their comparison. Understanding to what extent different fingerprints are contributing to the results could allow for precise control of model development. Therefore, the aim of this study is to compare the performance of models based on different types of molecular fingerprints in order to estimate which information they contain.