Current Bioinformatics - Volume 18, Issue 10, 2023

Background: The expression of secretory proteins is involved in each stage of biomass from fetal development to the immune response. As an animal model for the study of human diseases, the study of protein secretion in pigs has strong application prospects.Objective: Although secretory proteins play an important role in cell activities, there are no machine learning-based approaches for the prediction of pig secretory proteins. This study aims to establish a prediction model for identifying the secretory protein in Sus scrofa.Methods: Based on the pseudo composition of k-spaced amino acid pairs feature encoding method and support vector machine algorithm, a prediction model was established for the identification of the secretory protein in Sus scrofa.Results: The model produced the AUROC of 0.885 and 0.728 on the training set and independent testing set, respectively. In addition, we discussed features used for the prediction.Conclusion: In this study, we proposed the first classification model to identify secretory proteins in Sus scrofa. By learning the characteristic of secretory proteins, it may become feasible to design and produce secretory proteins with distinctive properties that are currently unavailable.

Introduction: It is expected that certain driver mutations may alter the gene expression of their associated or interacting partners, including cognate proteins.Methods: We introduced DEGdriver, a novel method that can discriminate between mutations in drivers and passengers by utilizing gene differential expression at the individual level.Results: After being tested on eleven TCGA cancer datasets, DEGdriver substantially outperformed cutting-edge approaches in distinguishing driver genes from passengers and exhibited robustness to varying parameters and protein-protein interaction networks.Conclusion: Through enrichment analysis, we prove that DEGdriver can identify functional modules or pathways in addition to novel driver genes.

Background: Understanding the subcellular location of proteins is essential for studying molecular and protein functions. Intracellular proteins must interact with appropriate molecules at the right time and in the right subcellular location to fulfill their functions. Therefore, the precise prediction of protein subcellular location can help elucidate complex cellular physiological response processes and is of great importance to research human diseases and pathophysiology.Methods: Traditional approaches to protein subcellular pattern analysis are primarily based on feature concatenation and classifier design. However, highly complex structures and poor performance are prominent shortcomings of these traditional approaches. In this paper, we report the development of an end-to-end pixel-enlightened neural network (IDRnet) based on Interactive Pointwise Attention (IPA) for the prediction of protein subcellular locations using immunohistochemistry (IHC) images. Patch splitting was adopted to reduce interference caused by tissue microarrays, such as bubbles, edges, and blanks. The IPA unit was constructed with a Depthwise and Pointwise convolution (DP) unit, and a pointwise pixel-enlightened algorithm was applied to modify and enrich protein subcellular location information.Results: IDRnet was able to achieve 97.33% accuracy in single-label IHC patch images and 88.59% subset accuracy in mixed-label IHC patch images, and outperformed other mainstream deep learning models. In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) was adopted to visualize the spatial information of proteins in the feature map, which helped to explain and understand the IHC image's abstract features and concrete expression form.

Introduction: Numerous clinical data and animal models demonstrate that many patients with neuropathic pain suffer from concomitant depressive symptoms.Methods: Massive evidence from biological experiments has verified that the medial prefrontal cortex (mPFC), prefrontal cortex, hippocampus, and other brain regions play an influential role in the comorbidity of neuropathic pain and depression, but the mechanism by which neuropathic pain induces depression remains unclear.Results: In this study, we mined existing publicly available databases of high-throughput sequencing data intending to identify the differentially expressed genes (DEGs) in the process of neuropathic paininduced depression.Conclusion: This study provides a rudimentary exploration of the mechanism of neuropathic paininduced depression and provides credible evidence for its management and precaution.

Background: Spinocerebellar ataxia type 3/Machado-Joseph disease (SCA3/MJD) is a clinically heterogeneous and progressive condition. Evaluation of its progression will contribute to clinical management and genetic counseling.Objective: The objective of this study was to provide a visualized interpretable prediction of the progression of SCA3/MJD based on machine learning (ML) methods.Methods: A total of 716 patients with SCA3/MJD were included in this study. The International Cooperative Ataxia Rating Scale (ICARS) and Scale for the Assessment and Rating of Ataxia (SARA) scores were used to quantitatively assess disease progression in the patients. Clinical and genotype information were collected as factors for predicting progression. Prediction models were constructed with ML algorithms, and the prediction results were then visualized to facilitate personalizing of clinical consultation.Results: The CAG repeat length of ATXN3 and its product with age, the duration of disease, and age were identified as the 4 most important factors for predicting the severity and progression of SCA3/MJD. The SVM-based model achieved the best performance in predicting the total ICARS and SARA scores, with accuracy (10%) values of 0.7619 for the SARA and 0.7042 for the ICARS. To visualize the predictions, line charts were used to show the expected progression over the next decade, and radar charts were used to show the scores of each part of the ICARS and SARA separately.Conclusion: We are the first group to apply ML algorithms to predict progression in SCA3/MJD and achieved desirable results. Visualization provided personalized predictions for each sample and can aid in developing clinical counseling regimens in the future.

Background: Prostate cancer (PCa) is a kind of malignant tumor with high incidence among males worldwide. The identification of novel biomarker signatures is therefore of clinical significance for PCa precision medicine. It has been acknowledged that the breaking of stability and vulnerability in biological network provides important clues for cancer biomarker discovery.Methods: In this study, a bioinformatics model by characterizing the centrality of nodes in PCa-specific protein-protein interaction (PPI) network was proposed and applied to identify novel gene signatures for PCa detection. Compared with traditional methods, this model integrated degree, closeness and betweenness centrality as the criterion for Hub gene prioritization. The identified biomarkers were validated based on receiver-operating characteristic evaluation, qRT-PCR experimental analysis and literatureguided functional survey.Results: Four genes, i.e., MYOF, RBFOX3, OCLN, and CDKN1C, were screened with average AUC ranging from 0.79 to 0.87 in the predicted and validated datasets for PCa diagnosis. Among them, MYOF, RBFOX3, and CDKN1C were observed to be down-regulated whereas OCLN was over-expressed in PCa groups. The in vitro qRT-PCR experiment using cell line samples convinced the potential of identified genes as novel biomarkers for PCa detection. Biological process and pathway enrichment analysis suggested the underlying role of identified biomarkers in mediating PCa-related genes and pathways including TGF-β, Hippo, MAPK signaling during PCa occurrence and progression.Conclusion: Novel gene signatures were screened as candidate biomarkers for PCa detection based on topological characterization of PCa-specific PPI network. More clinical validation using human samples will be performed in future work.

Background: New drugs are costly, time-consuming, and often accompanied by safety concerns. With the development of deep learning, computer-aided drug design has become more mainstream, and convolutional neural networks and graph neural networks have been widely used for drug128;“ target affinity (DTA) prediction.Objective: The paper proposes a method of predicting DTA using graph convolutional networks and multiscale convolutional neural networks.Methods: We construct drug molecules into graph representation vectors and learn feature expressions through graph attention networks and graph convolutional networks. A three-branch convolutional neural network learns the local and global features of protein sequences, and the two feature representations are merged into a regression module to predict the DTA.Results: We present a novel model to predict DTA, with a 2.5% improvement in the consistency index and a 21% accuracy improvement in terms of the mean squared error on the Davis dataset compared to DeepDTA. Morever, our method outperformed other mainstream DTA prediction models namely, GANsDTA, WideDTA, GraphDTA and DeepAffinity.Conclusion: The results showed that the use of multiscale convolutional neural networks was better than a single-branched convolutional neural network at capturing protein signatures and the use of graphs to express drug molecules yielded better results.