Drug discovery aims to keep fueling new medicines to cure and palliate many ailments and some untreatable diseases that still afflict humanity. The ADME/Tox (absorption, distribution, metabolism, excretion/toxicity) properties of candidate drug molecules are key factors that determine the safety, uptake, elimination, metabolic behavior and effectiveness of drug research and development. The predictive technique of ADME/Tox drastically reduces the fraction of pharmaceutics-related failure in the early stages of drug development. Driven by the expectation of accelerated timelines, reduced costs and the potential to reveal hidden insights from vast datasets, artificial intelligence techniques such as Graphormer are showing increasing promise and usefulness to perform custom models for molecule modeling tasks. However, Graphormer and other transformer-based models do not consider the molecular fingerprint, as well as the physicochemicals that have been proved effective in traditional computational drug research. Here, we propose an enhanced model based on Graphormer which uses a tree model that fully integrates some known information and achieves better prediction and interpretability. More importantly, the model achieves new state-of-the-art results on ADME/Tox properties prediction benchmarks, surpassing several challenging models. Experimental results demonstrate an average SMAPE (Symmetric Mean Absolute Percentage Error) of 18.9 and a PCC (Pearson Correlation Coefficient) of 0.86 on ADME/Tox prediction test sets. These findings highlight the efficacy of our approach and its potential to enhance drug discovery processes. By leveraging the strengths of Graphormer and incorporating additional molecular descriptors, our model offers improved predictive capabilities, thus contributing to the advancement of ADME/Tox prediction in drug development. The integration of various information sources further enables better interpretability, aiding researchers in understanding the underlying factors influencing the predictions. Overall, our work demonstrates the potential of our enhanced model to expedite drug discovery, reduce costs, and enhance the success rate of our pharmaceutical development efforts.

The table recognition model rows and columns aggregated network (RCANet) uses a semantic segmentation approach to recognize table structure, and achieves better performance in table row and column segmentation. However, this model uses ResNet18 as the backbone network, and the model has 11.35 million parameters and a volume of 45.5 M, which is inconvenient to deploy to lightweight servers or mobile terminals. Therefore, from the perspective of model compression, this paper proposes the lightweight rows and columns attention aggregated network (LRCAANet), which uses the lightweight network ShuffleNetv2 to replace the original RCANet backbone network ResNet18 to simplify the model size. Considering that the lightweight network reduces the number of feature channels, it has a certain impact on the performance of the model. In order to strengthen the learning between feature channels, the rows attention aggregated (RAA) module and the columns attention aggregated (CAA) module are proposed. The RAA module and the CAA module add the squeeze and excitation (SE) module to the original row and column aggregated modules, respectively. Adding the SE module means the model can learn the correlation between channels and improve the prediction effect of the lightweight model. The experimental results show that our method greatly reduces the model parameters and model volume while ensuring low-performance loss. In the end, the average F1 score of our model is only 1.77% lower than the original model, the parameters are only 0.17 million, and the volume is only 0.8 M. Compared with the original model, the parameter amount and volume are reduced by more than 95%.