Automated X-ray threat detection is challenged by cluttered baggage scans, severe object occlusions, and the scarcity of annotated datasets. Traditional supervised approaches are impractical, as they require large amounts of labeled data, which is difficult to obtain given the rarity of threat objects. Unsupervised methods, on the other hand, often fail to differentiate between threat and nonthreat items due to the complex grayscale nature and high object overlap inherent in X-ray imagery. To overcome these limitations, we propose X-SSL a novel self-supervised learning framework that eliminates manual annotations designed to perform threat localization. Our approach integrates spatial region extraction using MaskCut for zero-shot object proposal generation, contrastive multi-modal clustering that leverages both image and text encoders to cluster and label proposals into threat and nonthreat categories, and self-supervised knowledge distillation where a teacher–student model refines multiscale features from global and local image crops for improved representation learning. We evaluated X-SSL on two benchmark datasets: PIDray (39,000+ images) and CLCXray (14,000+ images), demonstrating significant improvements over previous state-of-the-art (SOTA) methods. In the hidden PIDray subset, X-SSL improves detection AP to 18.16 (+4.41 AP) and segmentation AP to 13.76 (+1.01 AP) over previous methods. On CLCXray, it achieves 39.26 detection AP (+5.93 AP) and 38.67 segmentation AP (+10.20 AP), significantly surpassing previous approaches. For classification, X-SSL achieves an accuracy of 43% on PIDray Hidden and 65% on CLCXray, further highlighting its superior performance compared to existing weakly supervised and unsupervised methods. Code will be available here: https://github.com/yonathan-kiflom/X-SSL .

DNA damage is a critical factor in the onset and progression of cancer. When DNA is damaged, the number of genetic mutations increases, making it necessary to activate DNA repair mechanisms. A crucial factor in the base excision repair process, which helps maintain the stability of the genome, is an enzyme called DNA polymerase β (Pol β) encoded by the POLB gene. It plays a vital role in the repair of damaged DNA. Additionally, variations known as Single Nucleotide Polymorphisms (SNPs) in the POLB gene can potentially affect the ability to repair DNA. This study uses bioinformatics tools that extract important features from SNPs to construct a feature matrix, which is then used in combination with machine learning algorithms to predict the likelihood of developing cancer associated with a specific mutation. Eight different machine learning algorithms were used to investigate the relationship between POLB gene variations and their potential role in cancer onset. This study not only highlights the complex link between POLB gene SNPs and cancer, but also underscores the effectiveness of machine learning approaches in genomic studies, paving the way for advanced predictive models in genetic and cancer research.

BACKGROUND: Plummeting DNA sequencing cost in recent years has enabled genome sequencing projects to scale up by several orders of magnitude, which is transforming genomics into a highly data-intensive field of research. This development provides the much needed statistical power required for genotype-phenotype predictions in complex diseases. METHODS: In order to efficiently leverage the wealth of information, we here assessed several genomic data science tools. The rationale to focus on on-premise installations is to cope with situations where data confidentiality and compliance regulations etc. rule out cloud based solutions. We established a comprehensive qualitative and quantitative comparison between BCFtools, SnpSift, Hail, GEMINI, and OpenCGA. The tools were compared in terms of data storage technology, query speed, scalability, annotation, data manipulation, visualization, data output representation, and availability. RESULTS: Tools that leverage sophisticated data structures are noted as the most suitable for large-scale projects in varying degrees of scalability in comparison to flat-file manipulation (e.g., BCFtools, and SnpSift). Remarkably, for small to mid-size projects, even lightweight relational database. CONCLUSION: The assessment criteria provide insights into the typical questions posed in scalable genomics and serve as guidance for the development of scalable computational infrastructure in genomics.

BACKGROUND: For some understudied populations, genotype data is minimal for genotype-phenotype prediction. However, we can use the data of some other large populations to learn about the disease-causing SNPs and use that knowledge for the genotype-phenotype prediction of small populations. This manuscript illustrated that transfer learning is applicable for genotype data and genotype-phenotype prediction. RESULTS: Using HAPGEN2 and PhenotypeSimulator, we generated eight phenotypes for 500 cases/500 controls (CEU, large population) and 100 cases/100 controls (YRI, small populations). We considered 5 (4 phenotypes) and 10 (4 phenotypes) different risk SNPs for each phenotype to evaluate the proposed method. The improved accuracy with transfer learning for eight different phenotypes was between 2 and 14.2 percent. The two-tailed p-value between the classification accuracies for all phenotypes without transfer learning and with transfer learning was 0.0306 for five risk SNPs phenotypes and 0.0478 for ten risk SNPs phenotypes. CONCLUSION: The proposed pipeline is used to transfer knowledge for the case/control classification of the small population. In addition, we argue that this method can also be used in the realm of endangered species and personalized medicine. If the large population data is extensive compared to small population data, expect transfer learning results to improve significantly. We show that Transfer learning is capable to create powerful models for genotype-phenotype predictions in large, well-studied populations and fine-tune these models to populations were data is sparse.

Scientific research is shedding light on the interaction of the gut microbiome with the human host and on its role in human health. Existing machine learning methods have shown great potential in discriminating healthy from diseased microbiome states. Most of them leverage shotgun metagenomic sequencing to extract gut microbial species-relative abundances or strain-level markers. Each of these gut microbial profiling modalities showed diagnostic potential when tested separately; however, no existing approach combines them in a single predictive framework. Here, we propose the Multimodal Variational Information Bottleneck (MVIB), a novel deep learning model capable of learning a joint representation of multiple heterogeneous data modalities. MVIB achieves competitive classification performance while being faster than existing methods. Additionally, MVIB offers interpretable results. Our model adopts an information theoretic interpretation of deep neural networks and computes a joint stochastic encoding of different input data modalities. We use MVIB to predict whether human hosts are affected by a certain disease by jointly analysing gut microbial species-relative abundances and strain-level markers. MVIB is evaluated on human gut metagenomic samples from 11 publicly available disease cohorts covering 6 different diseases. We achieve high performance (0.80 < ROC AUC < 0.95) on 5 cohorts and at least medium performance on the remaining ones. We adopt a saliency technique to interpret the output of MVIB and identify the most relevant microbial species and strain-level markers to the model's predictions. We also perform cross-study generalisation experiments, where we train and test MVIB on different cohorts of the same disease, and overall we achieve comparable results to the baseline approach, i.e. the Random Forest. Further, we evaluate our model by adding metabolomic data derived from mass spectrometry as a third input modality. Our method is scalable with respect to input data modalities and has an average training time of < 1.4 seconds. The source code and the datasets used in this work are publicly available.

BACKGROUND: Genotype-phenotype predictions are of great importance in genetics. These predictions can help to find genetic mutations causing variations in human beings. There are many approaches for finding the association which can be broadly categorized into two classes, statistical techniques, and machine learning. Statistical techniques are good for finding the actual SNPs causing variation where Machine Learning techniques are good where we just want to classify the people into different categories. In this article, we examined the Eye-color and Type-2 diabetes phenotype. The proposed technique is a hybrid approach consisting of some parts from statistical techniques and remaining from Machine learning. RESULTS: The main dataset for Eye-color phenotype consists of 806 people. 404 people have Blue-Green eyes where 402 people have Brown eyes. After preprocessing we generated 8 different datasets, containing different numbers of SNPs, using the mutation difference and thresholding at individual SNP. We calculated three types of mutation at each SNP no mutation, partial mutation, and full mutation. After that data is transformed for machine learning algorithms. We used about 9 classifiers, RandomForest, Extreme Gradient boosting, ANN, LSTM, GRU, BILSTM, 1DCNN, ensembles of ANN, and ensembles of LSTM which gave the best accuracy of 0.91, 0.9286, 0.945, 0.94, 0.94, 0.92, 0.95, and 0.96% respectively. Stacked ensembles of LSTM outperformed other algorithms for 1560 SNPs with an overall accuracy of 0.96, AUC = 0.98 for brown eyes, and AUC = 0.97 for Blue-Green eyes. The main dataset for Type-2 diabetes consists of 107 people where 30 people are classified as cases and 74 people as controls. We used different linear threshold to find the optimal number of SNPs for classification. The final model gave an accuracy of 0.97%. CONCLUSION: Genotype-phenotype predictions are very useful especially in forensic. These predictions can help to identify SNP variant association with traits and diseases. Given more datasets, machine learning model predictions can be increased. Moreover, the non-linearity in the Machine learning model and the combination of SNPs Mutations while training the model increases the prediction. We considered binary classification problems but the proposed approach can be extended to multi-class classification.

An amendment to this paper has been published and can be accessed via the original article.

BACKGROUND: What is a healthy microbiome? The pursuit of this and many related questions, especially in light of the recently recognized microbial component in a wide range of diseases has sparked a surge in metagenomic studies. They are often not simply attributable to a single pathogen but rather are the result of complex ecological processes. Relatedly, the increasing DNA sequencing depth and number of samples in metagenomic case-control studies enabled the applicability of powerful statistical methods, e.g. Machine Learning approaches. For the latter, the feature space is typically shaped by the relative abundances of operational taxonomic units, as determined by cost-effective phylogenetic marker gene profiles. While a substantial body of microbiome/microbiota research involves unsupervised and supervised Machine Learning, very little attention has been put on feature selection and engineering. RESULTS: We here propose the first algorithm to exploit phylogenetic hierarchy (i.e. an all-encompassing taxonomy) in feature engineering for microbiota classification. The rationale is to exploit the often mono- or oligophyletic distribution of relevant (but hidden) traits by virtue of taxonomic abstraction. The algorithm is embedded in a comprehensive microbiota classification pipeline, which we applied to a diverse range of datasets, distinguishing healthy from diseased microbiota samples. CONCLUSION: We demonstrate substantial improvements over the state-of-the-art microbiota classification tools in terms of classification accuracy, regardless of the actual Machine Learning technique while using drastically reduced feature spaces. Moreover, generalized features bear great explanatory value: they provide a concise description of conditions and thus help to provide pathophysiological insights. Indeed, the automatically and reproducibly derived features are consistent with previously published domain expert analyses.

BACKGROUND: Given the current influx of 16S rRNA profiles of microbiota samples, it is conceivable that large amounts of them eventually are available for search, comparison and contextualization with respect to novel samples. This process facilitates the identification of similar compositional features in microbiota elsewhere and therefore can help to understand driving factors for microbial community assembly. RESULTS: We present Visibiome, a microbiome search engine that can perform exhaustive, phylogeny based similarity search and contextualization of user-provided samples against a comprehensive dataset of 16S rRNA profiles environments, while tackling several computational challenges. In order to scale to high demands, we developed a distributed system that combines web framework technology, task queueing and scheduling, cloud computing and a dedicated database server. To further ensure speed and efficiency, we have deployed Nearest Neighbor search algorithms, capable of sublinear searches in high-dimensional metric spaces in combination with an optimized Earth Mover Distance based implementation of weighted UniFrac. The search also incorporates pairwise (adaptive) rarefaction and optionally, 16S rRNA copy number correction. The result of a query microbiome sample is the contextualization against a comprehensive database of microbiome samples from a diverse range of environments, visualized through a rich set of interactive figures and diagrams, including barchart-based compositional comparisons and ranking of the closest matches in the database. CONCLUSIONS: Visibiome is a convenient, scalable and efficient framework to search microbiomes against a comprehensive database of environmental samples. The search engine leverages a popular but computationally expensive, phylogeny based distance metric, while providing numerous advantages over the current state of the art tool.

Comprehensive mapping of environmental microbiomes in terms of their compositional features remains a great challenge in understanding the microbial biosphere of the Earth. It bears promise to identify the driving forces behind the observed community patterns and whether community assembly happens deterministically. Advances in Next Generation Sequencing allow large community profiling studies, exceeding sequencing data output of conventional methods in scale by orders of magnitude. However, appropriate collection systems are still in a nascent state. We here present a database of 20,427 diverse environmental 16S rRNA profiles from 2,426 independent studies, which forms the foundation of our meta-analysis. We conducted a sample size adaptive all-against-all beta diversity comparison while also respecting phylogenetic relationships of Operational Taxonomic Units(OTUs). After conventional hierarchical clustering we systematically test for enrichment of Environmental Ontology terms and their abstractions in all possible clusters. This post-hoc algorithm provides a novel formalism that quantifies to what extend compositional and semantic similarity of microbial community samples coincide. We automatically visualize significantly enriched subclusters on a comprehensive dendrogram of microbial communities. As a result we obtain the hitherto most differentiated and comprehensive view on global patterns of microbial community diversity. We observe strong clusterability of microbial communities in ecosystems such as human/mammal-associated, geothermal, fresh water, plant-associated, soils and rhizosphere microbiomes, whereas hypersaline and anthropogenic samples are less homogeneous. Moreover, saline samples appear less cohesive in terms of compositional properties than previously reported.

BACKGROUND: A large proportion of an organism's genome encodes for membrane proteins. Membrane proteins are important for many cellular processes, and several diseases can be linked to mutations in them. With the tremendous growth of sequence data, there is an increasing need to reliably identify membrane proteins from sequence, to functionally annotate them, and to correctly predict their topology. RESULTS: We introduce a technique called structural fragment clustering, which learns sequential motifs from 3D structural fragments. From over 500,000 fragments, we obtain 213 statistically significant, non-redundant, and novel motifs that are highly specific to alpha-helical transmembrane proteins. From these 213 motifs, 58 of them were assigned to function and checked in the scientific literature for a biological assessment. Seventy percent of the motifs are found in co-factor, ligand, and ion binding sites, 30% at protein interaction interfaces, and 12% bind specific lipids such as glycerol or cardiolipins. The vast majority of motifs (94%) appear across evolutionarily unrelated families, highlighting the modularity of functional design in membrane proteins. We describe three novel motifs in detail: (1) a dimer interface motif found in voltage-gated chloride channels, (2) a proton transfer motif found in heme-copper oxidases, and (3) a convergently evolved interface helix motif found in an aspartate symporter, a serine protease, and cytochrome b. CONCLUSIONS: Our findings suggest that functional modules exist in membrane proteins, and that they occur in completely different evolutionary contexts and cover different binding sites. Structural fragment clustering allows us to link sequence motifs to function through clusters of structural fragments. The sequence motifs can be applied to identify and characterize membrane proteins in novel genomes.

BACKGROUND: Many protein sequences are still poorly annotated. Functional characterization of a protein is often improved by the identification of its interaction partners. Here, we aim to predict protein-protein interactions (PPI) and protein-ligand interactions (PLI) on sequence level using 3D information. To this end, we use machine learning to compile sequential segments that constitute structural features of an interaction site into one profile Hidden Markov Model descriptor. The resulting collection of descriptors can be used to screen sequence databases in order to predict functional sites. RESULTS: We generate descriptors for 740 classified types of protein-protein binding sites and for more than 3,000 protein-ligand binding sites. Cross validation reveals that two thirds of the PPI descriptors are sufficiently conserved and significant enough to be used for binding site recognition. We further validate 230 PPIs that were extracted from the literature, where we additionally identify the interface residues. Finally we test ligand-binding descriptors for the case of ATP. From sequences with Swiss-Prot annotation "ATP-binding", we achieve a recall of 25% with a precision of 89%, whereas Prosite's P-loop motif recognizes an equal amount of hits at the expense of a much higher number of false positives (precision: 57%). Our method yields 771 hits with a precision of 96% that were not previously picked up by any Prosite-pattern. CONCLUSION: The automatically generated descriptors are a useful complement to known Prosite/InterPro motifs. They serve to predict protein-protein as well as protein-ligand interactions along with their binding site residues for proteins where merely sequence information is available.

MOTIVATION: Much research has been devoted to the characterization of interaction interfaces found in complexes with known structure. In this context, the interactions of non-homologous domains at equivalent binding sites are of particular interest, as they can reveal convergently evolved interface motifs. Such motifs are an important source of information to formulate rules for interaction specificity and to design ligands based on the common features shared among diverse partners. RESULTS: We develop a novel method to identify non-homologous structural domains which bind at equivalent sites when interacting with a common partner. We systematically apply this method to all pairs of interactions with known structure and derive a comprehensive database for these interactions. Of all non-homologous domains, which bind with a common interaction partner, 4.2% use the same interface of the common interaction partner (excluding immunoglobulins and proteases). This rises to 16% if immunoglobulin and proteases are included. We demonstrate two applications of our database: first, the systematic screening for viral protein interfaces, which can mimic native interfaces and thus interfere; and second, structural motifs in enzymes and its inhibitors. We highlight several cases of virus protein mimicry: viral M3 protein interferes with a chemokine dimer interface. The virus has evolved the motif SVSPLP, which mimics the native SSDTTP motif. A second example is the regulatory factor Nef in HIV which can mimic a kinase when interacting with SH3. Among others the virus has evolved the kinase's PxxP motif. Further, we elucidate motif resemblances in Baculovirus p35 and HIV capsid proteins. Finally, chymotrypsin is subject to scrutiny wrt. its structural similarity to subtilisin and wrt. its inhibitor's similar recognition sites. SUPPLEMENTARY INFORMATION: A database is online at scoppi.biotec.tu-dresden.de/abac/.

A systematic classification of protein-protein interfaces is a valuable resource for understanding the principles of molecular recognition and for modelling protein complexes. Here, we present a classification of domain interfaces according to their geometry. Our new algorithm uses a hybrid approach of both sequential and structural features. The accuracy is evaluated on a hand-curated dataset of 416 interfaces. Our hybrid procedure achieves 83% precision and 95% recall, which improves the earlier sequence-based method by 5% on both terms. We classify virtually all domain interfaces of known structure, which results in nearly 6,000 distinct types of interfaces. In 40% of the cases, the interacting domain families associate in multiple orientations, suggesting that all the possible binding orientations need to be explored for modelling multidomain proteins and protein complexes. In general, hub proteins are shown to use distinct surface regions (multiple faces) for interactions with different partners. Our classification provides a convenient framework to query genuine gene fusion, which conserves binding orientation in both fused and separate forms. The result suggests that the binding orientations are not conserved in at least one-third of the gene fusion cases detected by a conventional sequence similarity search. We show that any evolutionary analysis on interfaces can be skewed by multiple binding orientations and multiple interaction partners. The taxonomic distribution of interface types suggests that ancient interfaces common to the three major kingdoms of life are enriched by symmetric homodimers. The classification results are online at http://www.scoppi.org.