• Developed an explainable machine learning approach for transparent T2DM diagnosis. • Ensures both global and local model interpretability. • Provides patient-specific prediction explanations. • Extracted clinical insights align with existing medical knowledge. • Using data from two hospitals to validate the model internally and externally. The interpretability of disease prediction models is often crucial for their trustworthiness and usability among medical practitioners. Existing methods in interpretable artificial intelligence improve model transparency but fall short in identifying precise, disease-specific primal information. In this work, an interpretable deep learning-based algorithm called the data space landmark refiner was developed, which not only enhances both global interpretability and local interpretability but also reveals the intrinsic information of the data distribution. Using the proposed method, a type 2 diabetes mellitus diagnostic model with high interpretability was constructed on the basis of the electronic health records from two hospitals. Moreover, effective diagnostic information was directly derived from the model’s internal parameters, demonstrating strong alignment with current clinical knowledge. Compared with conventional interpretable machine learning approaches, the proposed method offered more precise and specific interpretability, increasing clinical practitioners’ trust in machine learning-supported diagnostic models.

As an important part of malware detection and classification, sequence-based analysis can be integrated into dynamic detection system for real-time detection. This work presents a novel learning method for malware detection models that leverages advances in graph embedding for fusing the n-gram data into a one-hot feature space with different transmission directions. By capturing the information flow, our method finds a better feature representation for detection tasks with rely solely on sequence information. To enhance the stability of feature representation, this work adopts a multi-task learning strategy which achieves better performance in independent testing. We evaluate our method on two different realworld datasets and compare it against four superior malware detection models. During malware detection using our method, we conducted in-depth discussions on feature length, graph embedding direction, model depth, and different multi-task learning strategies. Experimental and discussion results show that our method significantly outperforms alternative approaches across evaluation settings.

Predicting RNA solvent accessibility using only primary sequence data can be regarded as sequence-based prediction work. Currently, the established studies for sequence-based RNA solvent accessibility prediction are limited due to the available number of datasets and black box prediction. To improve these issues, we first expanded the available RNA structures and then developed a sequence-based model using modified attention layers with different receptive fields to conform to the stem-loop structure of RNA chains. We measured the improvement with an extended dataset and further explored the model's interpretability by analysing the model structures, attention values and hyperparameters. Finally, we found that the developed model regarded the pieces of a sequence as templates during the training process. This work will be helpful for researchers who would like to build RNA attribute prediction models using deep learning in the future.

MOTIVATION: Various bacterial pathogens can deliver their secreted substrates also called effectors through Type III secretion systems (T3SSs) into host cells and cause diseases. Since T3SS secreted effectors (T3SEs) play important roles in pathogen-host interactions, identifying them is crucial to our understanding of the pathogenic mechanisms of T3SSs. However, the effectors display high level of sequence diversity, therefore making the identification a difficult process. There is a need to develop a novel and effective method to screen and select putative novel effectors from bacterial genomes that can be validated by a smaller number of key experiments. RESULTS: We develop a deep convolution neural network to directly classify any protein sequence into T3SEs or non-T3SEs, which is useful for both effector prediction and the study of sequence-function relationship. Different from traditional machine learning-based methods, our method automatically extracts T3SE-related features from a protein N-terminal sequence of 100 residues and maps it to the T3SEs space. We train and test our method on the datasets curated from 16 species, yielding an average classification accuracy of 83.7% in the 5-fold cross-validation and an accuracy of 92.6% for the test set. Moreover, when comparing with known state-of-the-art prediction methods, the accuracy of our method is 6.31-20.73% higher than previous methods on a common independent dataset. Besides, we visualize the convolutional kernels and successfully identify the key features of T3SEs, which contain important signal information for secretion. Finally, some effectors reported in the literature are used to further demonstrate the application of DeepT3. AVAILABILITY AND IMPLEMENTATION: DeepT3 is freely available at: https://github.com/lje00006/DeepT3. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: CRISPR-Cpf1 has recently been reported as another RNA-guided endonuclease of class 2 CRISPR-Cas system, which expands the molecular biology toolkit for genome editing. However, most of the online tools and applications to date have been developed primarily for the Cas9. There are a limited number of tools available for the Cpf1. RESULTS: We present DeepCpf1, a deep convolution neural networks (CNN) approach to predict Cpf1 guide RNAs on-target activity and off-target effects using their matched and mismatched DNA sequences. Trained on published data sets, DeepCpf1 is superior to other machine learning algorithms and reliably predicts the most efficient and less off-target effects guide RNAs for a given gene. Combined with a permutation importance analysis, the key features of guide RNA sequences are identified, which determine the activity and specificity of genome editing. CONCLUSIONS: DeepCpf1 can significantly improve the accuracy of Cpf1-based genome editing and facilitates the generation of optimized guide RNAs libraries.