Structure-Aware Multimodal Learning Improves Minor-Class Signal Peptide Prediction.

Signal peptides are N-terminal presequences responsible for targeting proteins to the endomembrane system, and subsequent subcellular or extracellular compartments, and consequently condition their proper function. The significance of signal peptides stimulates development of new computational methods for their detection. These methods employ learning systems trained on datasets comprising signal peptides from different types of proteins and taxonomic groups. As a result, the accuracy of predictions are high in the case of signal peptides that are well-represented in databases, but might be low in other, atypical cases. Such atypical signal peptides are present in proteins found in apicomplexan parasites, causative agents of malaria and toxoplasmosis. Apicomplexan proteins have a unique amino acid composition due to their AT-biased genomes. Therefore, we designed a new, more flexible and universal probabilistic model for recognition of atypical eukaryotic signal peptides. Our approach called signalHsmm includes knowledge about the structure of signal peptides and physicochemical properties of amino acids. It is able to recognize signal peptides from the malaria parasites and related species more accurately than popular programs. Moreover, it is still universal enough to provide prediction of other signal peptides on par with the best preforming predictors.

Fast and effective prediction of signal peptides and their cleavage sites is of great importance in computational biology. There are two kinds of approaches developed to predict signal peptides, one of which based on model training approach such as SignalP and SPEPlip, and another based on sliding window method such as PrediSi, Signal-CF and Signal-3L. In this paper, the scaled window method proposed by Chou was employed to extract cleavable secretory segments, and template matching fusion method (named as Signal-TMF) was introduced to predict signal peptide cleavage sites. Comparing with Chou's Signal-CF method, the best overall accuracy of Signal-TMF is 6-15% higher than of Chou's in jackknife test. The Signal-TMF can also be used to effectively predict the long signal peptide cleavage site. The results show that Signal-TMF will be very useful to the areas related to signal peptides.

Automatic image-based disease severity estimation generally uses discrete (i.e., quantized) severity labels. Annotating discrete labels is often difficult due to the images with ambiguous severity. An easier alternative is to use relative annotation, which compares the severity level between image pairs. By using a learning-to-rank framework with relative annotation, we can train a neural network that estimates rank scores that are relative to severity levels. However, the relative annotation for all possible pairs is prohibitive, and therefore, appropriate sample pair selection is mandatory. This paper proposes a deep Bayesian active-learning-to-rank, which trains a Bayesian convolutional neural network while automatically selecting appropriate pairs for relative annotation. We confirmed the efficiency of the proposed method through experiments on endoscopic images of ulcerative colitis. In addition, we confirmed that our method is useful even with the severe class imbalance because of its ability to select samples from minor classes automatically.KeywordsComputer-aided diagnosisLearning to rankActive learningRelative annotationEndoscopic image dataset

Proteins destined for secretion or membrane compartments possess signal peptides for insertion into the membrane. The signal peptide is therefore critical for localization and function of cell surface receptors and ligands that mediate cell-cell communication. About 4% of all human proteins listed in UniProt database have signal peptide domains in their N terminals. A comprehensive literature survey was performed to retrieve functional and disease associated genetic variants in the signal peptide domains of human proteins. In 21 human proteins we have identified 26 disease associated mutations within their signal peptide domains, 14 mutations of which have been experimentally shown to impair the signal peptide function and thus influence protein transportation. We took advantage of SignalP 3.0 predictions to characterize the signal peptide prediction score differences between the mutant and the wild-type alleles of each mutation, as well as 189 previously uncharacterized single nucleotide polymorphisms (SNPs) found to be located in the signal peptide domains of 165 human proteins. Comparisons of signal peptide prediction outcomes of mutations and SNPs, have implicated SNPs potentially impacting the signal peptide function, and thus the cellular localization of the human proteins. The majority of the top candidate proteins represented membrane and secreted proteins that are associated with molecular transport, cell signaling and cell to cell interaction processes of the cell. This is the first study that systematically characterizes genetic variation occurring in the signal peptides of all human proteins. This study represents a useful strategy for prioritization of SNPs occurring within the signal peptide domains of human proteins. Functional evaluation of candidates identified herein may reveal effects on major cellular processes including immune cell function, cell recognition and adhesion, and signal transduction.

BackgroundProtein production as secretory-form is a powerful tool in industrial enzyme production due to the simple purification procedure. Streptomyces lividans is a versatile host for secretory production of useful proteins. In order to expand the amount of secreted protein, signal peptide sequences, which encourage protein secretion from inside cell to extracellular environment, are one of the most significant factors. In this study, we focused on Streptomyces lividans as a host strain to secrete useful proteins, and screened for signal peptides from the biomass-degradation enzymes derived from Thermobifida fusca YX and S. lividans.ResultsThree candidate signal peptides were isolated and evaluated for their protein secretion ability using β-glucosidase derived from T. fusca YX, which is a non-secreted protein, as a model protein. Using S. lividans xylanase C signal peptide, the amount of produced the β-glucosidase reached 10 times as much as that when using Streptomyces cinnamoneus phospholipase D signal peptide, which was identified as a versatile signal peptide in our previous report. In addition, the introduction of the β-glucosidase fused to xylanase C signal peptide using two kinds of plasmid, pUC702 and pTYM18, led to further protein secretion, and the maximal level of produced the β-glucosidase increased up to 17 times (1.1 g/l) compared to using only pUC702 carrying the β-glucosidase fused to S. cinnamoneus phospholipase D signal peptide.ConclusionIn the present study, we focused on signal peptide sequences derived from biomass degradation enzymes, which are usually secreted into the culture supernatant, and screened for signal peptides leading to effective protein secretion. Using the signal peptides, the hyper-protein secretion system was successfully demonstrated for the cytoplasmic β-glucosidase.

The corticotropin-releasing factor receptor type 2a (CRF(2(a)) receptor) belongs to the family of G protein-coupled receptors. The receptor possesses a putative N-terminal signal peptide that is believed to be cleaved-off after mediating the endoplasmic reticulum targeting/insertion process, like the corresponding sequence of the homologous CRF(1) receptor. Here, we have assessed the functional significance of the putative signal peptide of the CRF(2(a)) receptor and show that it is surprisingly completely incapable of mediating endoplasmic reticulum targeting, despite meeting all sequence criteria for a functional signal by prediction algorithms. Moreover, it is uncleaved and forms part of the mature receptor protein. Replacement of residue Asn(13) by hydrophobic or positively charged residues converts the sequence into a fully functional and cleaved signal peptide demonstrating that conventional signal peptide functions are inhibited by a single amino acid residue. Deletion of the domain leads to an increase in the amount of immature, intracellularly retained receptors demonstrating that the sequence has adopted a new function in receptor trafficking through the early secretory pathway. Taken together, our results identify a novel hydrophobic receptor domain in the family of the heptahelical G protein-coupled receptors and the first pseudo signal peptide of a eukaryotic membrane protein. Our data also show that the extreme N termini of the individual CRF receptor subtypes differ substantially.

As next-generation sequence (NGS) production continues to increase, analysis is becoming a significant bottleneck. However, in situations where information is required only for specific sequence variants, it is not necessary to assemble or align whole genome data sets in their entirety. Rather, NGS data sets can be mined for the presence of sequence variants of interest by localized assembly, which is a faster, easier, and more accurate approach. We present TASR, a streamlined assembler that interrogates very large NGS data sets for the presence of specific variants, by only considering reads within the sequence space of input target sequences provided by the user. The NGS data set is searched for reads with an exact match to all possible short words within the target sequence, and these reads are then assembled strin-gently to generate a consensus of the target and flanking sequence. Typically, variants of a particular locus are provided as different target sequences, and the presence of the variant in the data set being interrogated is revealed by a successful assembly outcome. However, TASR can also be used to find unknown sequences that flank a given target. We demonstrate that TASR has utility in finding or confirming ge-nomic mutations, polymorphism, fusion and integration events. Targeted assembly is a powerful method for interrogating large data sets for the presence of sequence variants of interest. TASR is a fast, flexible and easy to use tool for targeted assembly.

The genes coding for human and mouse AMACO, an extracellular matrix protein containing VWA-like domains related to those in MAtrilins and COllagens, were detected in databases, the cDNAs were cloned, and the primary structures were deduced from the nucleotide sequences. The genes consist of 14 exons and have a similar exon/intron organization. The protein consists of a signal peptide sequence, an N-terminal VWA domain connected to two additional, tandem VWA domains by a cysteine-rich sequence and an epidermal growth factor (EGF)-like domain. The C terminus is made up of another EGF-like domain followed by a unique sequence present in mouse, but absent in human. The predicted molecular weight of the proteins is 79,485 in human and 83,024 in mouse. Full-length AMACO was expressed in 293-EBNA cells, purified by use of an affinity tag and subjected to biochemical characterization. Both monomers and aggregates of AMACO were recovered, as shown by electron microscopy and SDS-PAGE. AMACO was found in the media of a variety of established cell lines of both fibroblast and epithelial origin. In the matrix formed by 293-EBNA cells overexpressing the protein, AMACO was deposited in patchy structures that were often cell-associated. Affinity-purified antibodies detect expression in cartilage and expression associated with certain basement membranes. In the kidney of adult mice, a second promoter located in intron 4 is active. If the resulting transcript is translated it could not yield a secreted protein because of the lack of a signal peptide sequence. The developmental switch from an AMACO mRNA, expressed by the newborn kidney, to the truncated transcript found in the adult kidney indicates an unusual regulation of AMACO expression.

Objective To predict the secretome and analyze the secreted proteins and signal peptides of Echinococcus multilocularis using genome-wide bioinformatics approaches thus to provide a platform for finding biomarkers for development of diagnosis and drugs.Methods Signal peptides of E.multilocularis were identified from the whole genome sequence using SignalP4.1 program,and the proteins containing signal sequences were analyzed with TMHMM v2.0,Phobius,Big-PI predictor and TargetPl.1 in a stepwise way to minimize the false-positive prediction.Subsequently,the sequence features of both the signal peptides and the secreted proteins were statistically analyzed by SPSS19.0 and Excel.Differences in the numbers of amino acid of each secreted and the non-secreted sequences were determined through kolmogorov-smirnov test (K-S test)and t test.Finally,the KAAS (KEGG Automatic Annotation Sever) functional annotation and clustering were performed for the secretome sequences.Results A total of 875 proteins encoding sequences containing signal peptides was found in 10 780 E.multilocularis genome sequences.Among them,307 sequences are membrane-binding proteins,38 proteins contain GPI anchor site and 12 proteins are located in mitochondria.Finally,a total of 518 proteins was recognized as secreted proteins.The signal sequences contain mostly 11-53amino acids,among them 61％ are hydrophobic.The secreted proteins identified possess 38-7 809 amino acids,which is significantly less than the amount possessed by non-secreted proteins(11-11 194 amino acids) (t=0.203,P＜0.01).KAAS analysis showed that these secreted proteins are mainly involved in human disease,metabolism,environmental information processing,organismal systems,cellular process and genetic information processing.Among them,6 sequences are related to parasitic infection.Conclusion The secretome of E.multilocularis containing 518 secreted protein sequences was predicted and analyzed thus to provide a database for further identification of diagnostics,vaccine and drug targets. Key words: Echinococcus multilocularis; Secretome; Signal peptide; Bioinformatics

Graph convolutional neural networks have difficulties in model training takes a lot of time and lack of universality in meta-path design when processing knowledge graphs. To solve these issues, this paper proposes an efficient recommendation algorithm integrating knowledge graphs with graph convolutional networks(EKG-GCN). The algorithm is mainly based on the graph convolution algorithm. Firstly, the relationship between users and entities is standardized and scored, then a new neighborhood aggregation method is proposed. Some nodes in the knowledge graph are set as central nodes. A distance-influence function is set up according to the distance from other nodes to the central node, and the feature extraction of information in the knowledge graph is completed according to the distance-influence function. Finally, the knowledge graph features are used as an auxiliary to complete the recommendation task. In the experiment, three data sets were selected to compare with other advanced models, and it was found that the model is in an advantageous position in terms of training time and recommendation effect, and has a certain degree of universality.

Bacterial proteins with cleaved or uncleaved signal peptides of the general secretory pathway

Limited experimental data remains a key challenge in applying machine learning to drug discovery, particularly for cancer-related targets. In this study, we present a data-efficient active meta-deep learning framework to predict mitogen-activated protein kinase 1 (MAPK1) inhibitors, which are promising candidates for cancer-related therapies. Our approach integrates active learning (AL) with a meta-model that combines four deep architectures: a convolutional neural network, an attention, a graph convolutional network, and a graph neural network-attention, trained on molecular descriptors and graph-based representations. These models generate four probability-based features that feed into an attention-based meta-learner, improving predictive performance by 5.12% in the area under the precision-recall curve (AUPRC) and 5.48% in the Matthews correlation coefficient (MCC) using only 10% of the training data. Among the AL sampling strategies evaluated, entropy sampling showed competitive performance in selecting informative molecules for model improvement. Overall, our framework achieves an AUPRC of 0.835 ± 0.017 and MCC of 0.817 ± 0.017, on par with a traditional training method despite using only 26.7% of the training data. Compared to a conventional random forest model trained on brute-force, a 100% full training set, our approach shows a 10.6% improvement in AUPRC and modest gains in MCC, confirming the effectiveness of the proposed framework. Under severe class imbalance, balanced accuracy steadily increased across AL iterations, reaching values greater than 0.85 at the final iteration for all uncertainty-driven strategies. Molecular docking confirmed successful prioritization of the top four predicted compounds. Evaluation on an external MAPK1 data set demonstrated generalizability, with our approach achieving an AUPRC of 0.818 and an MCC of 0.403, comparable to the independent test set. These results highlight the potential of combining intelligent data selection with deep learning architectures through the meta-model to accelerate predictive performance in data-scarce drug discovery. Scientific contribution: This study contributes a novel, data-efficient active meta-deep learning framework for predicting MAPK1 inhibitors, addressing the challenge of limited experimental data in a cancer-specific target. By integrating AL with a meta-model composed of four deep architectures, the approach significantly enhances the predictive performance using only a fraction of the training data. The framework achieves superior metrics compared to traditional training methods, highlighting its potential to accelerate drug discovery in data-scarce settings.

Translocation of a nascent protein from the cytosol into the ER mediated by its signal peptide is a critical step in protein secretion. The aim of this work was to develop a platform technology to optimize the signal peptides for high level production of therapeutic antibodies in CHO cells. A database of signal peptides from a large number of human immunoglobulin (Ig) heavy chain (HC) and kappa light chain (LC) was generated. Most of the HC signal peptides contain 19 amino acids which can be divided into three domains and the LC signal peptides contain 22 amino acids. The signal peptides were then clustered according to sequence similarity. Based on the clustering, 8 HC and 2 LC signal peptides were analyzed for their impacts on the production of 5-top selling antibody therapeutics, namely, Herceptin, Avastin, Remicade, Rituxan, and Humira. The best HC and LC signal peptides for producing these 5 antibodies were identified. The optimized signal peptides for Rituxan is 2-fold better compared to its native signal peptides which are available in the public database. Substitution of a single amino acid in the optimized HC signal peptide for Avastin reduced its production significantly. Mass spectrometry analyses revealed that all optimized signal peptides are accurately removed in the mature antibodies. The results presented in this report are particularly important for the production of these 5 antibodies as biosimilar drugs. They also have the potential to be the best signal peptides for the production of new antibodies in CHO cells.

ECG databases are usually highly imbalanced due to the abundance of Normal ECG and scarcity of abnormal cases. As such, deep learning classifiers trained on imbalanced datasets usually perform poorly, especially on minor classes. One solution is to generate realistic synthetic ECG signals using Generative Adversarial Networks (GAN) to augment imbalanced datasets. In this study, we combined conditional GAN with WGAN-GP and developed AC-WGAN-GP in 1D form for the first time to be applied on MIT-BIH Arrhythmia dataset. We investigated the impact of data augmentation on arrhythmia classification. Two models were employed for ECG generation: (i) unconditional GAN; Wasserstein GAN with gradient penalty (WGAN-GP) is trained on each class individually; (ii) conditional GAN; one Auxiliary Classifier WGAN-GP (AC-WGAN-GP) model is trained on all classes and then used to generate synthetic beats in all classes. Two scenarios are defined for each case: (a) unscreened; all the generated synthetic beats were used, and (b) screened; only high-quality beats are selected and used, based on their Dynamic Time Warping (DTW) distance to a designated template. The state-of-the-art ResNet classifier (EcgResNet34) is trained on each of the four augmented datasets and the performance metrics (precision/recall/F1-Score micro- and macro-averaged, confusion matrices, multiclass precision-recall curves) were compared with those of the original imbalanced case. We also used a simple metric Net Improvement. All the three metrics show consistently that unconditional GAN with raw generated data creates the best improvements.

Although hundreds of different signal peptides have now been identified, few studies have examined the factors enabling signal peptides to augment secretion of mature proteins. Signal peptides, located at the N-terminus of nascent secreted proteins, characteristically have three domains: (1) a basic domain at the N-terminus, (2) a central hydrophobic core, and (3) a carboxy-terminal cleavage region. In this study, we investigated whether alterations in the basic and/or the hydrophobic domains of a commonly used signal peptide from interleukin-2 (IL-2) affected secretion of two proteins: placental alkaline phosphatase (AP) and endostatin. A series of modifications in the basic and/or hydrophobic domains of the IL-2 signal peptide were made by polymerase chain reaction with endostatin or AP plasmids as templates. Transfection of wild-type or modified IL-2 signal peptides fused in-frame with endostatin or AP were done with Superfect in vitro or by the hydrodynamic method in vivo. Increasing both the basicity and hydrophobicity of the signal peptide augmented the secretion of AP and endostatin by approximately 2.5- and 3.5-fold, respectively, from MDA-MB-435 cells in vitro. Over a range of DNA concentrations and times, the most effective IL-2 signal peptide increased AP levels in the medium compared to the wild-type IL-2 signal peptide. Comparable results from these modified IL-2 signal peptides were found to increase AP levels in the medium from bovine aortic endothelial cells. Moreover, the combined changes in basic and hydrophobic domains of the IL-2 signal peptide augmented serum levels of endostatin and placental AP by 3-fold when the optimal plasmid constructs were injected in vivo. Modification of the IL-2 signal peptide augments protein secretion both in vitro and in vivo. As a result, optimizing the signal peptide should be considered for increasing the therapeutic levels of secreted proteins.