“pySiRC”: Machine Learning Combined with Molecular Fingerprints to Predict the Reaction Rate Constant of the Radical-Based Oxidation Processes of Aqueous Organic Contaminants

Abstract

We developed a web application structured in a machine learning and molecular fingerprint algorithm for the automatic calculation of the reaction rate constant of the oxidative processes of organic pollutants by • OH and SO 4•− radicals in the aqueous phasethe pySiRC platform. The model development followed the OECD principles: internal and external validation, applicability domain, and mechanistic interpretation. Three machine learning algorithms combined with molecular fingerprints were evaluated, and all the models resulted in high goodness-of-fit for the training set with R 2 > 0.931 for the • OH radical and R 2 > 0.916 for the SO 4•− radical and good predictive capacity for the test set with R ext 2 = Q ext 2 values in the range of 0.639−0.823 and 0.767−0.824 for the • OH and SO 4•− radicals. The model was interpreted using the SHAP (SHapley Additive exPlanations) method: the results showed that the model developed made the prediction based on a reasonable understanding of how electron-withdrawing and -donating groups interfere with the reactivity of the • OH and SO 4•− radicals. We hope that our models and web interface can stimulate and expand the application and interpretation of kinetic research on contaminants in water treatment units based on advanced oxidative technologies.

“pySewage”: a hybrid approach to predict the number of SARS-CoV-2-infected people from wastewater in Brazil

“…Bioinformatics tools play an important role in monitoring SARS-CoV-2 and provide non-computational users with the possibility to analyze data and advanced knowledge related to COVID-19 (Hufsky et al 2021 ). This is probably part of the reason that although several studies have been conducted to monitor SARS-CoV-2, and estimate the number of infected people from wastewater samples, they are mostly used by a small group of researchers (Sanches-Neto et al 2021 ). Therefore, in addition to studies of SARS-CoV-2 monitoring by experimental techniques such as RT-qPCR and genomic sequencing, it is extremely important to develop web applications to automate the SARS-CoV-2 detection combined with COVID-19 monitoring (Pérez-Cataluña et al 2022 ).…”

Transfer Learning Approach to Multitarget Temperature-Dependent Reaction Rate Prediction

“…Mean and max Tanimoto similarity between compounds in the test set and the training set were used to assess the applicability domain as proposed in ref . The chosen threshold to include or remove a molecule from the applicability domain was found by incrementally varying the Tanimoto similarity from 0 to 0.4 and 0.05 for the max and mean, respectively.…”

“…Methods like SHAP (SHapley Additive exPlanations, a game theoretic approach to explain outputs of machine learning models) could be applied to add physical meaning and intuition. This was applied recently for rate constant predictions where important variables from fixed fingerprints were identified . However, this is not possible in our case since we have learned representations that do not exactly correspond to physical features.…”

“…We also attempt to quantify the applicability domain in a manner similar to that presented in ref . We use the continuous variable form of the Tanimoto coefficient to assess the similarity between two compounds, A and B , defined by where S is the similarity and x ji is the j th feature of molecule i .…”

“…More recently, ANNs were used to predict temperature-dependent site-specific rates for alkanes + OH reactions; however, the study was limited to only 11 reactions . Machine learning models on fixed fingerprints were also used for the prediction of rate constants of aqueous OH and SO 4 reactions. , Performance of such models on small data sets could be enhanced by combining small data sets and transferring knowledge as recently shown for SO 4 , HClO, O 3 , and ClO 2 . Fixed fingerprints, like the ones used in these studies (e.g., Morgan and MACCS), are widely used in quantitative structure property/activity relationship (QSPR/QSAR) studies and are easily generated through open-source software which makes them preferable compared to expertly crafted molecular descriptors.…”