Machine learning prediction of UV–Vis spectra features of organic compounds related to photoreactive potential

Abstract

Machine learning (ML) algorithms were explored for the classification of the UV–Vis absorption spectrum of organic molecules based on molecular descriptors and fingerprints generated from 2D chemical structures. Training and test data (~ 75 k molecules and associated UV–Vis data) were assembled from a database with lists of experimental absorption maxima. They were labeled with positive class (related to photoreactive potential) if an absorption maximum is reported in the range between 290 and 700 nm (UV/Vis) with molar extinction coefficient (MEC) above 1000 Lmol−1 cm−1, and as negative if no such a peak is in the list. Random forests were selected among several algorithms. The models were validated with two external test sets comprising 998 organic molecules, obtaining a global accuracy up to 0.89, sensitivity of 0.90 and specificity of 0.88. The ML output (UV–Vis spectrum class) was explored as a predictor of the 3T3 NRU phototoxicity in vitro assay for a set of 43 molecules. Comparable results were observed with the classification directly based on experimental UV–Vis data in the same format.

Phototoxicity─Medicinal Chemistry Strategies for Risk Mitigation in Drug Discovery

“…Recently, Mamede et al 34 reported their work of applying several state-of-the-art machine learning technologies in the classification of UV−vis spectra related to the photoreactive potential of organic compounds. Based on the ICH S10 guidance on photosafety evaluation of pharmaceuticals, compounds were labeled as positive class if they were able to absorb light between 290 and 700 nm (UV−vis range) with a molar extinction coefficient (MEC) greater than 1000 L•mol −1 • cm −1 ; otherwise, they were labeled as negative class.…”

Data-Driven Prediction of Circular Dichroism-Based Calibration Curves for the Rapid Screening of Chiral Primary Amine Enantiomeric Excess Values

“…16 However, examples in which spectroscopic properties associated with analytical methods for screening are modeled and predicted are less common. 17 For ee assays that rely on CD, the signal is largely governed by differences in the steric size of the groups on the analyte's point stereocenter. 18 Our group has used this steric dependence to predict the calibration curves of a tripodal zinc(II) complex for the ee determination of chiral secondary alcohols, as well as a tripodal quinoline-Cu(II) sensor for chiral carboxylic acids.…”

“…Recently, the use of multiparameter LFERs has found wide success in modeling reaction outcomes in asymmetric catalysis by guiding reaction discovery and optimization via screening . However, examples in which spectroscopic properties associated with analytical methods for screening are modeled and predicted are less common …”

Functional Output Regression for Machine Learning in Materials Science

“…With these applications, we demonstrate the potential predictive ability of the proposed methods on small amounts of training data. We compare ordinary regression, which predicts a scalar output variable with a pre-quantified spectral feature, with the present methods predicting the whole function directly, and show the superiority of the latter and its statistical mechanisms in relation to multitask learning. , The Python codes used in the case studies were distributed…”