As Android is a popular a mobile operating system, Android malware is on the rise, which poses a great threat to user privacy and security. Considering the poor detection effects of the single feature selection algorithm and the low detection efficiency of traditional machine learning methods, we propose an Android malware detection framework based on stacking ensemble learning—MFDroid—to identify Android malware. In this paper, we used seven feature selection algorithms to select permissions, API calls, and opcodes, and then merged the results of each feature selection algorithm to obtain a new feature set. Subsequently, we used this to train the base learner, and set the logical regression as a meta-classifier, to learn the implicit information from the output of base learners and obtain the classification results. After the evaluation, the F1-score of MFDroid reached 96.0%. Finally, we analyzed each type of feature to identify the differences between malicious and benign applications. At the end of this paper, we present some general conclusions. In recent years, malicious applications and benign applications have been similar in terms of permission requests. In other words, the model of training, only with permission, can no longer effectively or efficiently distinguish malicious applications from benign applications.

In recent years, Android malware has continued to evolve against detection technologies, becoming more concealed and harmful, making it difficult for existing models to resist adversarial sample attacks. At the current stage, the detection result is no longer the only criterion for evaluating the pros and cons of the model with its algorithms, it is also vital to take the model’s defensive ability against adversarial samples into consideration. In this study, we propose a general framework named AdvAndMal, which consists of a two-layer network for adversarial training to generate adversarial samples and improve the effectiveness of the classifiers in Android malware detection and family classification. The adversarial sample generation layer is composed of a conditional generative adversarial network called pix2pix, which can generate malware variants to extend the classifiers’ training set, and the malware classification layer is trained by RGB image visualized from the sequence of system calls. To evaluate the adversarial training effect of the framework, we propose the robustness coefficient, a symmetric interval i = [−1, 1], and conduct controlled experiments on the dataset to measure the robustness of the overall framework for the adversarial training. Experimental results on 12 families with the largest number of samples in the Drebin dataset show that the accuracy of the overall framework is increased from 0.976 to 0.989, and its robustness coefficient is increased from 0.857 to 0.917, which proves the effectiveness of the adversarial training method.

The integrity of cloud-based convolutional neural network (CNN) prediction services can be jeopardized by a malicious cloud server. Although zero-knowledge proof approaches can be used to verify integrity, they are difficult to use for larger CNN models like LeNet-5 and VGG16, due to the large cost (in terms of time and storage) of generating a proof. This paper proposes ValidCNN, which can efficiently generate integrity proofs based zk-SNARK. At the heart of ValidCNN, it is a novel usage of Freivald's concepts for circuit construction, and a more efficient way for verifying matrix multiplication. Our experimental results demonstrate that VaildCNN significantly outperforms the state-of-the-art approaches that are based on zk-SNARK. For example, compared with ZEN, VaildCNN achieves a 12-fold improvement in time and a 31-fold improvement in storage. Compared with vCNN, VaildCNN achieves a 195-fold and 279-fold improvement in time and storage respectively.

Accurately identifying potential drug-target interactions (DTIs) is a critical step in drug discovery. Multiple heterogeneous biological data provide abundant features for DTI prediction. Many computational methods have been proposed based on these data. However, most of these methods either extract features from sequences or from networks, utilizing only one aspect of the characteristics of drugs and targets, neglecting the complementary information between these two types of features. In fact, integrating different types of features will provide more valuable information for DTI prediction. In this article, we propose a novel method to improve the predictive capability for DTIs, named MFCADTI, by integrating multi-source feature through cross-attention mechanisms. The method extracts network topological features from the heterogeneous network and attribute features from sequences of drugs and targets. Considering the complementarity and heterogeneity between network and attribute features, cross-attention mechanisms are used to integrate the network and attribute features of drugs and targets. To capture the correlations between drugs and targets, cross-attention is used to learn the interaction features of each drug-target pair. We evaluate MFCADTI on two datasets and experimental results demonstrate a significant improvement in the performance of MFCADTI compared to state-of-the-art methods. Finally, case studies illustrate that MFCADTI is an effective DTI prediction way that provides valuable guidance for drug development. The data and source code used in this study are available at: https://github.com/Dejavun/MFCADTI .

Model prediction based on machine learning is provided as a service in cloud environments, but how to verify that the model prediction service is entirely conducted becomes a critical challenge. Although zero-knowledge proof techniques potentially solve the integrity verification problem when applied to the prediction integrity of massive privacy-preserving Convolutional Neural Networks (CNNs), the significant proof burden results in low practicality. In this research, we present psvCNN (parallel splitting zero-knowledge technique for integrity verification). The psvCNN scheme effectively improves the utilization of computational resources in CNN prediction integrity, proving by an independent splitting design. Through a convolutional kernel-based model splitting design and an underlying zero-knowledge succinct non-interactive knowledge argument, our psvCNN develops parallelizable zero-knowledge proof circuits for CNN prediction. Furthermore, psvCNN presents an updated Freivalds algorithm for a faster integrity verification process. Experiments show that psvCNN is practical and efficient in terms of proof time and storage, generating a prediction integrity proof with a proof size of 1.2MB in 7.65s for the structurally complicated CNN model VGG16. psvCNN is 3765 times faster than the latest zk-SNARK-based non-interactive method vCNN, and 12 times faster than the latest sumcheck-based interactive technique zkCNN in terms of proving time.

Background: The identification of disease-related genes is of great significance for the diagnosis and treatment of human disease. Most studies have focused on developing efficient and accurate computational methods to predict disease-causing genes. Due to the sparsity and complexity of biomedical data, it is still a challenge to develop an effective multi-feature fusion model to identify disease genes. Results: This paper proposes an approach to predict the pathogenic gene based on multi-head attention fusion (MHAGP). Firstly, the heterogeneous biological information networks of disease genes are constructed by integrating multiple biomedical knowledge databases. Secondly, two graph representation learning algorithms are used to capture the feature vectors of gene-disease pairs from the network, and the features are fused by introducing multi-head attention. Finally, multi-layer perceptron model is used to predict the gene-disease association. Conclusions: The MHAGP model outperforms all of other methods in comparative experiments. Case studies also show that MHAGP is able to predict genes potentially associated with diseases. In the future, more biological entity association data, such as gene-drug, disease phenotype-gene ontology and so on, can be added to expand the information in heterogeneous biological networks and achieve more accurate predictions. In addition, MHAGP with strong expansibility can be used for potential tasks such as gene-drug association and drug-disease association prediction.