Holistic Prediction of the pK_a in Diverse Solvents Based on a Machine‐Learning Approach

Abstract

While many approaches to predict aqueous pK a values exist, the fast and accurate prediction of non-aqueous pK a values is still challenging. Based on the iBonD experimental pK a database (39 solvents), ah olistic pK a prediction model was established using machine learning.S tructural and physical-organic-parameter-based descriptors (SPOC) were introduced to represent the electronic and structural features of the molecules.The models trained with aneural network or the XGBoost algorithm showed the best prediction performance with alow MAE value of 0.87 pK a units.The approachallows ac omprehensive mapping of all possible pK a correlations between different solvents and it was validated by predicting the aqueous pK a and micro-pK a of pharmaceutical molecules and pK a values of organocatalysts in DMSO and MeCN with high accuracy.Anonline prediction platform was constructed based on the current model, which can providepK a prediction for different types of XÀHacidity in the most commonly used solvents.

Catalytic asymmetric synthesis of quaternary trifluoromethyl α- to ε-amino acid derivativesviaumpolung allylation/2-aza-Cope rearrangement

“…To our delight, with isatin-activated ketoimine ester 1c, the desired quaternary triuoromethyl a-amino acid derivative 3a was isolated in 99% yield with 96% ee. Considering the signicant effect of the isatin-derived imino moiety on the reactivity and the predicted pK a values (in DMSO: 17.38 (1a), 12.66 (1b), 12.44 (1c)), 18,19 we believed that both the steric effect and the electronic effect affect the reactivity of the tested ketoimine esters. Encouraged by these promising results, further screening of other parameters was performed.…”

Electron Rich Triarylphosphines as Nucleophilic Catalysts for Oxa-Michael Reactions

“…1. The benchmark catalyst TPP is unable to promote the oxa-Michael reaction of the good Michael acceptor 1 (electrophilicity parameter E of -19.05) 17 with the least acidic alcohol 2-propanol a as virtually no conversion was observed after 24 h. Using MMTPP leads to a minor improvement and 3 % conversion towards 1a was found after 24 h. TMTPP, however, gives already 4 % conversion after 1 h and 38 % conversion after 24 h. More acidic 1-propanol b reacts in the presence of TPP (27 % conversion after 24 h). MMTPP already provides a considerable improvement since a conversion of 66% is obtained after 24 h but TMTPP is again a distinctly better catalyst providing 73 % conversion after 1 h and almost full conversion (98 %) after 24 h. Allyl alcohol c is more reactive than 1-propanol as conversions with all catalysts at all conditions are slightly higher.…”

“…12 However, with base catalysis (KO t Bu) even better results than those presented here can be achieved. 20,21 Switching to the weaker Michael acceptor acrylamide (E=-23.54 for N,Ndimethylacrylamide), 17 no useful conversions on any account are obtained. However, TMTPP performs best, giving 61 and 74 % conversion with 1-propanol and allyl alcohol after 24 h. To illustrate that the reaction does not stop after 24 h the conversions were re-checked after 21 d. After this time with TMTPP as the catalyst, conversions of 44 % (3a), 92% (3b), 98 % (3c) and 91% (3d) are obtained.…”

“…Using activated potassium carbonate, a reaction temperature of 40 °C and 4 h reaction time give typically better conversions than those reported herein with nucleophiles. 23 Next, difunctional divinyl sulfone was tested as the strongest Michael acceptor (E = -18.36, for phenyl vinyl sulfone) 17 under investigation. In distinction from the experiments described above, three equivalents of the alcohol were used.…”

Machine Learning Enables Highly Accurate Predictions of Photophysical Properties of Organic Fluorescent Materials: Emission Wavelengths and Quantum Yields

“…Reported ML predictions are limited to Power Conversion Efficiency (PCE) [48][49][50][51][52] , gas absorption selectivity [53][54][55] and AIE effect 56 , most relying on expensive quantum mechanical calculations to generate input expressions. For solvated molecules, expression of solvent features is critical but rarely studied in detail 57 . To achieve large-scale predictions for emission wavelength and PLQY with low/no sacrifice in accuracy, new strategies for feature engineering as well as the selection/designing of ML algorithms must be explored.…”