Machine learning models for predicting endocrine disruption potential of environmental chemicals

Classification of estrogenic compounds by coupling high content analysis and machine learning algorithms

“…predicting the endocrine-disrupting potential of compounds on the ER. These holistic classification models allow us to predict the ER activity of compounds without running in separate agonist or antagonist modes and are shown to possess comparable classification performances with the previously reported values [6,13,16]. In addition, among 40 different experimental features studied in this work, "Array to Nucleoplasm Intensity Ratio" is found to be the top informative feature through a series of supervised and unsupervised analyses, and the results indicate that it is essential for predicting the ER activity of compounds through generalized predictive models.…”

“…Furthermore, recent efforts have also focused on coupling high throughput experimentation with computational methods for enabling the rapid diagnosis of the estrogenic potential of various chemicals via in silico predictions [6,[11][12][13][14][15][16]. Judson et al [6] used a linear model to predict the estrogenic activity of 1812 commercial and environmental chemicals based on the activity patterns across in vitro assays.…”

“…The authors also investigated the performance of 4 different classification models, namely k-nearest neighbors (kNN), local lazy method (lazy IB1), alternating decision tree (ADTree), and an integrated consensus model [14]. In another study by Chierici et al [16], deep learning and support vector machine (SVM) models were developed using the Collaborative Estrogen Receptor Activity Prediction Project (CERAPP) ToxCast data set for predicting the effects of EDCs on ER binding activity. A further detailed overview of in silico toxicity predictions using machine learning algorithms is provided in the notable review by Idakwo et al [17].…”

“…Furthermore, recent efforts have also focused on coupling high throughput experimentation with computational methods for enabling the rapid diagnosis of the estrogenic potential of various chemicals via in silico predictions [ 6 , 11 – 16 ]. Judson et al .…”

Realizing the promise of computational prediction in toxicology applications

“…Similarly, Chierici et al presented ML4Tox, a framework that enables deep learning and support vector machine prediction of not only binding but also agonist and antagonist interactions. 5 The authors applied the tool to the estrogen receptor's ligand binding domain, but the interested user could apply it to any target of interest. Helpfully, Chierici et al make their code available and Li et al make their web portal available for the community to use.…”