Jonny Proppe

The rapid growth of worldwide computing power has transformed in silico chemistry into a discipline that is integrated into the daily work of many chemists. Nowadays, researchers find it increasingly straightforward to predict a wide range of molecular properties and chemical processes at reasonable computational cost. The resulting abundance of data generated by quantum chemistry, molecular dynamics simulations, and chemical machine learning naturally raises questions about accuracy, precision, and reliability as well as the systematic treatment of errors and uncertainties. Addressing these questions through rigorous mathematical frameworks is at the heart of uncertainty quantification. In the past years, the incorporation of uncertainty quantification into in silico chemistry has gained attention, motivated by its ability to provide deeper insights into chemical phenomena. In this review, we establish a common language for uncertainty quantification with respect to in silico chemistry, introduce the key mathematical formalisms, and survey the growing body of work that applies uncertainty quantification across different areas of in silico chemistry.

Explainability methods in machine learning-driven research are increasingly being used, but it remains challenging to assess their reliability without deeply investigating the specific problem at hand.

In our previous work, we introduced WISP (Workflow for Interpretability Scoring using matched molecular Pairs), which enables users to quantitatively assess the performance of explainability methods for machine learning models. In this work, we focus on more complex tasks, such as yield prediction, pK_i values for inhibition of coagulation Factor Xa and AMES mutagenicity, where the explanations of the predicted property need to capture more intricate interaction patterns between structural motifs of either the reaction partner, the protein or DNA. Expanding upon part I of the “When Machine Learning Learns Chemistry” series, we demonstrate additional functionalities of the WISP workflow. Alongside the model and descriptor—agnostic atom attributor, WISP integrates a SHAP-based and an RDKit-based attribution method, enabling the comparison of multiple explainability approaches, as demonstrated on the Factor Xa dataset. This work also showcases WISP's capability to evaluate explanations for classification tasks such as AMES mutagenicity. The application of WISP to the AMES mutagenicity dataset reveals that the respective machine learning model fails to learn the underlying chemical relationships, instead relying primarily on numerical correlations. When applied to the yield dataset, WISP highlights specific cases where explainability methods that usually perform well fail to provide meaningful insights. This highlights WISP's ability to detect such limitations in trained models, providing valuable insights to guide targeted improvements in model development and data quality.

Base‐mediated C–H carboxylation is a versatile pathway for utilizing carbon dioxide (CO ₂ ) as a C1 building block in organic synthesis. However, CO ₂ constitutes a notorious thermodynamic sink, which restricts this approach to activated or intrinsically reactive nucleophiles. To qualitatively assess the stability of CO ₂ adducts, we present a computational approach that integrates quantum chemistry with statistical modeling to build a predictive workflow. The target property is the CO ₂ affinity, specifically the negative Gibbs free reaction energy. This predictive workflow has been applied to 60 novel carbon‐centered nucleophiles, suggesting reactions that yield stable carboxylation adducts. The results have been validated through experimental methods for five carbanions, which include three stable and two unstable adducts in DMSO according to our predictions. In addition, we examined two further carbanions that were suggested to form stable CO ₂ adducts in DMSO, to further assess the experimental protocol and broaden its scope to structurally distinct motifs.

Base‐mediated C–H carboxylation is a versatile pathway for utilizing carbon dioxide (CO ₂ ) as a C1 building block in organic synthesis. However, CO ₂ constitutes a notorious thermodynamic sink, which restricts this approach to activated or intrinsically reactive nucleophiles. To qualitatively assess the stability of CO ₂ adducts, we present a computational approach that integrates quantum chemistry with statistical modeling to build a predictive workflow. The target property is the CO ₂ affinity, specifically the negative Gibbs free reaction energy. This predictive workflow has been applied to 60 novel carbon‐centered nucleophiles, suggesting reactions that yield stable carboxylation adducts. The results have been validated through experimental methods for five carbanions, which include three stable and two unstable adducts in DMSO according to our predictions. In addition, we examined two further carbanions that were suggested to form stable CO ₂ adducts in DMSO, to further assess the experimental protocol and broaden its scope to structurally distinct motifs.

Explainability methods in machine learning-driven research are increasingly being used, but it remains challenging to assess their reliability without deeply investigating the specific problem at hand. In this work, we present a Python-based Workflow for Interpretability Scoring using matched molecular Pairs (WISP). This workflow can be applied to assess the performance of explainability methods on any given dataset containing SMILES and is model-agnostic, making it compatible with any machine learning model. Evaluation on two physics-based datasets demonstrates that the explanations reliably capture the predictions of the respective machine learning models. Furthermore, our workflow reveals that explainability methods can only meaningfully reflect the property of interest when the underlying models achieve high predictive accuracy. Therefore, the explainability performance on a test set can function as a quality measure of the underlying model. To ensure compatibility with any model type, we developed an atom attributor, which generates atom-level attributions for any model using any descriptor that can be obtained using SMILES representations. This method can also be applied as a standalone explainability tool, independently of WISP. WISP enables users to interpret a wide range of machine learning models in the chemical domain and gain valuable insights into how these models operate and the extent to which they capture underlying chemical principles.

In our previous work, we introduced WISP (Workflow for Interpretability Scoring using matched molecular Pairs), which enables users to quantitatively assess the performance of explainability methods for machine learning models. In this work, we focus on more complex tasks, such as yield prediction, pKi values for inhibition of coagulation Factor Xa and AMES mutagenicity, where the explanations of the predicted property need to capture more intricate interaction patterns between structural motifs of either the reaction partner, the protein or DNA. Expanding upon Part I of the "When Machine Learning Learns Chemistry" series, we demonstrate additional functionalities of the WISP workflow. Alongside the model and descriptor-agnostic atom attributor, WISP integrates a SHAP-based and an RDKit-based attribution method, enabling the comparison of multiple explainability approaches, as demonstrated on the Factor Xa dataset. This work also showcases WISP’s capability to evaluate explanations for classification tasks such as AMES mutagenicity. The application of WISP to the AMES mutagenicity dataset reveals that the respective machine learning model fails to learn the underlying chemical relationships, instead relying primarily on numerical correlations. When applied to the yield dataset, WISP highlights specific cases where explainability methods that usually perform well fail to provide meaningful insights. This highlights WISP’s ability to detect such limitations in trained models, providing valuable insights to guide targeted improvements in model development and data quality.

Rate constants for reactions between nucleophiles and electrophiles can be efficiently estimated by the Mayr–Patz equation, log k = s_N(N + E), which relies on three reactivity parameters as input. Utilising this equation, the Mayr group established a reference set for determining reactivity parameters of uncharged π-nucleophiles with positively charged electrophiles (benzhydrylium ions). Subsequently, the initial reference set was expanded by uncharged quinone methide electrophiles and carbanionic nucleophiles, resulting in an extension of the reactivity scales toward stronger nucleophiles and weaker electrophiles. For this work, the extended reference set was systematically analysed by automated algorithms to identify key aspects for future expansions. Lower bounds for reaction-to-species and reaction-to-parameter ratios were determined, ensuring minimal overfitting during parameter optimisation. It is also shown that the removal of electrophilic or nucleophilic species with high species-specific model errors in their predicted rate constants had positive results on the overall model error, with one species having a particularly great influence. These findings and the proposed methods may help future efforts to determine reliable data sets for the construction of Mayr-type reactivity scales.

Increasingly more research areas rely on machine learning methods to accelerate discovery while saving resources. Machine learning models, however, usually require large datasets of experimental or computational results, which in certain fields—such as (bio)chemistry, materials science, or medicine—are rarely given and often prohibitively expensive to obtain. To bypass that obstacle, active learning methods are employed to develop machine learning models with a desired performance while requiring the least possible number of computational or experimental results from the domain of application. For this purpose, the model’s knowledge about certain regions of the application domain is estimated to guide the choice of the model’s training set. Although active learning is widely studied for classification problems (discrete outcomes), comparatively few works handle this method for regression problems (continuous outcomes). In this work, we present our Python package regAL, which allows users to evaluate different active learning strategies for regression problems. With a minimal input of just the dataset in question, but many additional customization and insight options, this package is intended for anyone who aims to perform and understand active learning in their problem-specific scope. Program summary Program title: regAL¹ 1 regAL is an acronym for Active Learning of regression problems. When we speak German, however, we pronounce it as [] (meaning ‘shelf’ in German). Program source: https://doi.org/10.5281/zenodo.15309124, https://git.rz.tu-bs.de/proppe-group/active-learning/regAL Programming language: Python 3+ Program dependencies: numpy, scikit-learn, matplotlib, pandas

We present a data-driven approach that integrates supervised learning, quantum chemistry, and uncertainty quantification to determine CO₂ reactivity, enabling advances in carbon capture and the design of value-added chemicals.