There are currently 85,000 chemicals registered with the Environmental Protection Agency (EPA) under the Toxic Substances Control Act, but only a small fraction have measured toxicological data. To address this gap, high-throughput screening (HTS) and computational methods are vital. As part of one such HTS effort, embryonic zebrafish were used to examine a suite of morphological and mortality endpoints at six concentrations from over 1,000 unique chemicals found in the ToxCast library (phase 1 and 2). We hypothesized that by using a conditional generative adversarial network (cGAN) or deep neural networks (DNN), and leveraging this large set of toxicity data we could efficiently predict toxic outcomes of untested chemicals. Utilizing a novel method in this space, we converted the 3D structural information into a weighted set of points while retaining all information about the structure. In vivo toxicity and chemical data were used to train two neural network generators. The first was a DNN (Go-ZT) while the second utilized cGAN architecture (GAN-ZT) to train generators to produce toxicity data. Our results showed that Go-ZT significantly outperformed the cGAN, support vector machine, random forest and multilayer perceptron models in cross-validation, and when tested against an external test dataset. By combining both Go-ZT and GAN-ZT, our consensus model improved the SE, SP, PPV, and Kappa, to 71.4%, 95.9%, 71.4% and 0.673, respectively, resulting in an area under the receiver operating characteristic (AUROC) of 0.837. Considering their potential use as prescreening tools, these models could provide in vivo toxicity predictions and insight into the hundreds of thousands of untested chemicals to prioritize compounds for HT testing.
We extend the weighted-views molecular featurization framework (d’Avezac, Botts, Mohlenkamp, and Zunger, 2011) by enriching it with single-atom properties and Coulomb pair information. We use these enriched views within a machine-learning architecture incorporating shared Transformers whose attention scores directly incorporate Coulomb interactions, weighted aggregation of view outputs, and a final prediction neural network. We demonstrate performance competitive with the state-of-the-art on the QM7 atomization energy benchmark (regression), ranking enantiomers by their docking scores (trained as regression, tested by ranking), lipophilicity (regression), and blood–brain barrier penetration (classification).
6 October 2026 in SIAM Journal on Applied Dynamical Systems
Budding yeast has been observed for many years to exhibit stable metabolic oscillations in certain bioreactor experiments and some of these oscillations are known to involve the cell cycle and temporal clustering, i.e., groups of cells traversing the cycle in near synchrony. In budding yeast, division is asymmetric: it produces two nonidentical cells, namely a larger mother and a smaller daughter. A mother cell may be able to reach the next division more quickly than a daughter. Previous studies have suggested that this disparity may cause there to exist periodic temporally clustered solutions in which there are [Formula: see text] mother clusters and [Formula: see text] daughter clusters for [Formula: see text]. We study a population model of the cell cycle in which cells in one region of the cell cycle may negatively influence the progress of cells in another region, resulting in temporal clustering. We demonstrate numerically in simple simulations that [Formula: see text] periodic configurations can form spontaneously from a disordered population. As the main result, we prove analytically in the models that certain types of [Formula: see text] clustered solutions can exist and be linearly asymptotically stable.