Geometrical Interpretation and Design of Multilayer Perceptrons

QMGBP-DL: a deep learning and machine learning approach for quantum molecular graph band-gap prediction

Predicting molecular and quantum material properties, especially the band gap, is crucial for accelerating discoveries in drug design and material science. Although graph neural networks and probabilistic encoders are well established in molecular data analysis, their targeted integration and application for band-gap prediction remain an active research area. This paper introduces QMGBP-DL, a deep learning approach that combines a molecular graph encoder with machine learning models to improve the prediction accuracy of molecular and material band-gap energy. The encoder uses graph convolutional networks to derive latent representations of chemical structures from SMILES strings, optimized via Kullback-Leibler divergence loss. These representations serve as inputs for training various machine learning models to predict properties. QMGBP-DL's effectiveness is assessed using the QM9, PCQM4M, and OPV datasets, demonstrating significant improvements, particularly with a random forest model for property prediction. A comparative analysis against established approaches DenseGNN, MEGNet, and ALIGNN reveals that QMGBP-DL excels in predicting HOMO, LUMO, and band gap, achieving notably lower MAE values. The integration of GCN-derived latent spaces with traditional machine learning models, especially Random Forest, provides a powerful approach for band-gap prediction. The results highlight the efficacy of our integrated approach, showcasing that graph-based molecular encoding combined with machine learning, particularly Random Forest, is highly effective for accurate band-gap prediction, thereby facilitating material discovery and design.

CapsRule: Explainable Deep Learning for Classifying Network Attacks

Despite the potential deep learning (DL) algorithms have shown, their lack of transparency hinders their widespread application. Extracting if-then rules from deep neural networks is a powerful explanation method to capture nonlinear local behaviors. However, existing rule extraction methods suffer from inefficiency, incomprehensibility, infidelity, and not scaling well. Concerning security applications, they are not optimized regarding the decision boundary, data types and ranges, classification tasks, and dataset size. In this article, we propose CapsRule, an effective and efficient rule-based DL explanation method dedicated to classifying network attacks. It extracts high-fidelity rules from the feed-forward capsule network that explains how an input sample is classified. Using precomputed coupling coefficients, the training phase overlaps the rule extraction process to increase efficiency. The activation vector of a capsule can represent semantic intelligence about the attributes of the input sample. The rules extracted from CapsRule address the major concerns of network attack detection. The rules: 1) approximate the nonlinear decision boundary of the underlying data; 2) reduce the number of false positives significantly; 3) increase transparency; and 4) help find errors and noise in the data. We evaluate CapsRule on the CICDDoS2019 dataset that contains over a million of the most advanced Distributed Denial-of-Service (DDoS) attacks. The extensive evaluation shows that it generates accurate, high-fidelity, and comprehensible rules. CapsRule achieves an average accuracy of 99.0% and a false positive rate of 0.70% for reflection- and exploitation-based attacks. We verify that the learned features from the rulesets match our domain-specific knowledge. They also help find flaws in the dataset generation process and erroneous patterns caused by attack simulators.

Predicting Corynebacterium glutamicum promoters based on novel feature descriptor and feature selection technique

The promoter is an important noncoding DNA regulatory element, which combines with RNA polymerase to activate the expression of downstream genes. In industry, artificial arginine is mainly synthesized by Corynebacterium glutamicum. Replication of specific promoter regions can increase arginine production. Therefore, it is necessary to accurately locate the promoter in C. glutamicum. In the wet experiment, promoter identification depends on sigma factors and DNA splicing technology, this is a laborious job. To quickly and conveniently identify the promoters in C. glutamicum, we have developed a method based on novel feature representation and feature selection to complete this task, describing the DNA sequences through statistical parameters of multiple physicochemical properties, filtering redundant features by combining analysis of variance and hierarchical clustering, the prediction accuracy of the which is as high as 91.6%, the sensitivity of 91.9% can effectively identify promoters, and the specificity of 91.2% can accurately identify non-promoters. In addition, our model can correctly identify 181 promoters and 174 non-promoters among 400 independent samples, which proves that the developed prediction model has excellent robustness.