No abstract

Chemical structure extraction from documents remains a hard problem due to both false positive identification of structures during segmentation and errors in the predicted structures. Current approaches rely on handcrafted rules and subroutines that perform reasonably well generally, but still routinely encounter situations where recognition rates are not yet satisfactory and systematic improvement is challenging. Complications impacting performance of current approaches include the diversity in visual styles used by various software to render structures, the frequent use of ad hoc annotations, and other challenges related to image quality, including resolution and noise. We here present end-to-end deep learning solutions for both segmenting molecular structures from documents and for predicting chemical structures from these segmented images. This deep learning-based approach does not require any handcrafted features, is learned directly from data, and is robust against variations in image quality and style. Using the deep-learning approach described herein we show that it is possible to perform well on both segmentation and prediction of low resolution images containing moderately sized molecules found in journal articles and patents.

Evidence is emerging that elongation factor 2 kinase (eEF-2K) has potential as a target for anti-cancer therapy and possibly for the treatment of depression. Here the steady-state kinetic mechanism of eEF-2K is presented using a peptide substrate and is shown to conform to an ordered sequential mechanism with ATP binding first. Substrate inhibition by the peptide was observed and revealed to be competitive with ATP, explaining the observed ordered mechanism. Several small molecules are reported to inhibit eEF-2K activity with the most notable being the histidine kinase inhibitor NH125, which has been used in a number of studies to characterize eEF-2K activity in cells. While NH125 was previously reported to inhibit eEF-2K in vitro with an IC50 of 60 nM its mechanism of action was not established. Using the same kinetic assay the ability of an authentic sample of NH125 to inhibit eEF-2K was assessed over a range of substrate and inhibitor concentrations. A typical dose-response curve for the inhibition of eEF-2K by NH125 is best fit to an IC50 of 18 ± 0.25 µM and a Hill coefficient of 3.7 ± 0.14, suggesting that NH125 is a weak inhibitor of eEF-2K under the experimental conditions of a standard in vitro kinase assay. To test the possibility that NH125 is a potent inhibitor of eEF2 phosphorylation we assessed its ability to inhibit the phosphorylation of eEF2. Under standard kinase assay conditions NH125 exhibits similar weak ability to inhibit the phosphorylation of eEF2 by eEF-2K. Notably, the activity of NH125 is severely abrogated by the addition of 0.1 % triton to the kinase assay through a process that can be reversed upon dilution. These studies suggest that NH125 is as a non-specific colloidal aggregator in vitro, a notion further supported by the observation that NH125 inhibits other protein kinases, such as ERK2 and TRPM7 in a similar manner to eEF-2K. As NH125 is reported to inhibit eEF-2K in a cellular environment its ability to inhibit eEF2 phosphorylation was assessed in MDA-MB-231 breast cancer, A549 lung cancer and HEK-293T cell lines using a Western blot approach. No sign of decrease in the level of eEF2 phosphorylation was observed up to 12 hours following addition of NH125 to the media. Furthermore, contrary to the previously reported literatures, NH125 induced the phosphorylation of eEF-2.

Eukaryotic elongation factor 2 kinase (eEF-2K) is an atypical protein kinase regulated by Ca2+ and calmodulin (CaM). Its only known substrate is eukaryotic elongation factor 2 (eEF-2), whose phosphorylation by eEF-2K impedes global protein synthesis. To date, the mechanism of eEF-2K autophosphorylation has not been fully elucidated. To investigate the mechanism of autophosphorylation, human eEF-2K was co-expressed with λ-phosphatase, and purified from bacteria in a three-step protocol using a calmodulin-affinity column. Purified eEF-2K was induced to autophosphorylate by incubation with Ca2+/CaM in the presence of MgATP. Analyzing tryptic or chymotryptic peptides by mass spectrometry monitored the autophosphorylation over 0–180 minutes. The following five major autophosphorylation sites were identified, Thr-348, Thr-353, Ser-445, Ser-474 and Ser-500. In the presence of Ca2+/CaM, robust phosphorylation of Thr-348 occurs within seconds of adding MgATP. Mutagenesis studies suggest that phosphorylation of Thr-348 is required for substrate (eEF-2 or a peptide substrate) phosphorylation, but not self-phosphorylation. Phosphorylation of Ser-500 lags behind the phosphorylation of Thr-348, and is associated with calcium-independent activity of eEF-2K. Mutation of Ser-500 to Asp, but not Ala, renders eEF-2K calcium-independent. Surprisingly, this calcium-independent activity requires the presence of calmodulin.

Materials exhibiting higher mobilities than conventional organic semiconducting materials such as fullerenes and fused thiophenes are in high demand for applications in printed electronics. To discover new molecules in the heteroacene family that might show improved hole mobility, three de novo design methods were applied. Machine learning (ML) models were generated based on previously calculated hole reorganization energies of a quarter million examples of heteroacenes, where the energies were calculated by applying density functional theory (DFT) and a massive cloud computing environment. The three generative methods applied were (1) the continuous space method, where molecular structures are converted into continuous variables by applying the variational autoencoder/decoder technique; (2) the method based on reinforcement learning of SMILES strings (the REINVENT method); and (3) the junction tree variational autoencoder method that directly generates molecular graphs. Among the three methods, the second and third methods succeeded in obtaining chemical structures whose DFT-calculated hole reorganization energy was lower than the lowest energy in the training dataset. This suggests that an extrapolative materials design protocol can be developed by applying generative modeling to a quantitative structure−property relationship (QSPR) utility function.