Gene based message passing for drug repurposing

Do multisource data matter for NGP prediction? Evidence from the G-LSTM model

Precisely predicting natural gas prices (NGPs) is important because it can provide the necessary decision-making basis for energy scheduling, planning and control. However, NGPs are affected by many factors and exhibit the characteristics of nonlinearity and randomness, which makes accurate predictions challenging. Therefore, in this paper, the information gain of multisource data and the global optimization ability of the gray wolf algorithm are used to build a multifactor-driven NGP hybrid forecasting model to improve the prediction performance. First, the emotional tendency and readability of news text are extracted and calculated by using VADER and textstat tools, respectively. Then the network search index is filtered and integrated by using the correlation coefficient method and the CRITIC method to form alternative variables of multisource data (news and search index). Second, the gray wolf optimization algorithm is used to find and determine the best key parameter group in long short-term memory model. Finally, the spot price of natural gas in Henry Hub from March 1, 2012 to February 28, 2022 is selected as the prediction object, and multi-scenario numerical experiments are carried out to verify the effectiveness of the proposed model. The ablation experiment results show that the information gain brought by multisource data can effectively improve the prediction effect of NGPs. Furthermore, the proposed model has the best prediction performance in different scenarios and can be regarded as a promising prediction tool.

Predicting disease-gene associations through self-supervised mutual infomax graph convolution network

Illuminating associations between diseases and genes can help reveal the pathogenesis of syndromes and contribute to treatments, but a large number of associations remained unexplored. To identify novel disease-gene associations, many computational methods have been developed using disease and gene-related prior knowledge. However, these methods remain of relatively inferior performance due to the limited external data sources and the inevitable noise among the prior knowledge. In this study, we have developed a new method, Self-Supervised Mutual Infomax Graph Convolution Network (MiGCN), to predict disease-gene associations under the guidance of external disease-disease and gene-gene collaborative graphs. The noises within the collaborative graphs were eliminated by maximizing the mutual information between nodes and neighbors through a graphical mutual infomax layer. In parallel, the node interactions were strengthened by a novel informative message passing layer to improve the learning ability of graph neural network. The extensive experiments showed that our model achieved performance improvement over the state-of-art method by more than 8 % on AUC. The datasets, source codes and trained models of MiGCN are available at https://github.com/biomed-AI/MiGCN.

Generating Potential RET-Specific Inhibitors Using a Novel LSTM Encoder-Decoder Model

The receptor tyrosine kinase RET (rearranged during transfection) plays a vital role in various cell signaling pathways and is a critical factor in the development of the nervous system. Abnormal activation of the RET kinase can lead to several cancers, including thyroid cancer and non-small-cell lung cancer. However, most RET kinase inhibitors are multi-kinase inhibitors. Therefore, the development of an effective RET-specific inhibitor continues to present a significant challenge. To address this issue, we built a molecular generation model based on fragment-based drug design (FBDD) and a long short-term memory (LSTM) encoder-decoder structure to generate receptor-specific molecules with novel scaffolds. Remarkably, our model was trained with a molecular assembly accuracy of 98.4%. Leveraging the pre-trained model, we rapidly generated a RET-specific-candidate active-molecule library by transfer learning. Virtual screening based on our molecular generation model was performed, combined with molecular dynamics simulation and binding energy calculation, to discover specific RET inhibitors, and five novel molecules were selected. Further analyses indicated that two of these molecules have good binding affinities and synthesizability, exhibiting high selectivity. Overall, this investigation demonstrates the capacity of our model to generate novel receptor-specific molecules and provides a rapid method to discover potential drugs.