Abstract The tremendous success of deep learning for imaging applications has resulted in numerous beneficial advances. Unfortunately, this success has also been a catalyst for malicious uses such as photo-realistic face swapping of parties without consent. In this study, we use deep transfer learning for face swapping detection, showing true positive rates greater than 96% with very few false alarms. Distinguished from existing methods that only provide detection accuracy, we also provide uncertainty for each prediction, which is critical for trust in the deployment of such detection systems. Moreover, we provide a comparison to human subjects. To capture human recognition performance, we build a website to collect pairwise comparisons of images from human subjects. Based on these comparisons, we infer a consensus ranking from the image perceived as most real to the image perceived as most fake. Overall, the results show the effectiveness of our method. As part of this study, we create a novel dataset that is, to the best of our knowledge, the largest swapped face dataset created using still images. This dataset will be available for academic research use per request. Our goal of this study is to inspire more research in the field of image forensics through the creation of a dataset and initial analysis.

Identifying malicious software executables is made difficult by the constant adaptations introduced by miscreants in order to evade detection by antivirus software. Such changes are akin to mutations in biological sequences. Recently, high-throughput methods for gene sequence classification have been developed by the bioinformatics and computational biology communities. In this paper, we apply methods designed for gene sequencing to detect malware in a manner robust to attacker adaptations. Whereas most gene classification tools are optimized for and restricted to an alphabet of four letters (nucleic acids), we have selected the Strand gene sequence classifier for malware classification. Strand’s design can easily accommodate unstructured data with any alphabet, including source code or compiled machine code. To demonstrate that gene sequence classification tools are suitable for classifying malware, we apply Strand to approximately 500 GB of malware data provided by the Kaggle Microsoft Malware Classification Challenge (BIG 2015) used for predicting nine classes of polymorphic malware. Experiments show that, with minimal adaptation, the method achieves accuracy levels well above 95% requiring only a fraction of the training times used by the winning team’s method.

As more and more applications produce streaming data, clustering data streams has become an important technique for data and knowledge engineering. A typical approach is to summarize the data stream in real-time with an online process into a large number of so called micro-clusters. Micro-clusters represent local density estimates by aggregating the information of many data points in a defined area. On demand, a (modified) conventional clustering algorithm is used in a second offline step to recluster the micro-clusters into larger final clusters. For reclustering, the centers of the micro-clusters are used as pseudo points with the density estimates used as their weights. However, information about density in the area between micro-clusters is not preserved in the online process and reclustering is based on possibly inaccurate assumptions about the distribution of data within and between micro-clusters (e.g., uniform or Gaussian). This paper describes DBSTREAM, the first micro-cluster-based online clustering component that explicitly captures the density between micro-clusters via a shared density graph. The density information in this graph is then exploited for reclustering based on actual density between adjacent micro-clusters. We discuss the space and time complexity of maintaining the shared density graph. Experiments on a wide range of synthetic and real data sets highlight that using shared density improves clustering quality over other popular data stream clustering methods which require the creation of a larger number of smaller micro-clusters to achieve comparable results.

Gene Ontology is one of the largest bioinformatics project that seeks to consolidate knowledge about genes through annotation of terms to three ontologies. In this work, we present a technique to find association relationships in the annotation terms for the Saccharomyces cerevisiae (SGD) genome. We first present a normalization algorithm to ensure that the annotation terms have a similar level of specificity. Association rule mining algorithms are used to find significant and non-trivial association rules in these normalized datasets. Metrics such as support, confidence, and lift can be used to evaluate the strength of found rules. We conducted experiments on the entire SGD annotation dataset and here we present the top 10 strongest rules for each of the three ontologies. We verify the found rules using evidence from the biomedical literature. The presented method has a number of advantages - it relies only on the structure of the gene ontology, has minimal memory and storage requirements, and can be easily scaled for large genomes, such as the human genome. There are many applications of this technique, such as predicting the GO annotations for new genes or those that have not been studied extensively.

BACKGROUND: Next Generation Sequencing techniques are producing enormous amounts of biological sequence data and analysis becomes a major computational problem. Currently, most analysis, especially the identification of conserved regions, relies heavily on Multiple Sequence Alignment and its various heuristics such as progressive alignment, whose run time grows with the square of the number and the length of the aligned sequences and requires significant computational resources. In this work, we present a method to efficiently discover regions of high similarity across multiple sequences without performing expensive sequence alignment. The method is based on approximating edit distance between segments of sequences using p-mer frequency counts. Then, efficient high-throughput data stream clustering is used to group highly similar segments into so called quasi-alignments. Quasi-alignments have numerous applications such as identifying species and their taxonomic class from sequences, comparing sequences for similarities, and, as in this paper, discovering conserved regions across related sequences. RESULTS: In this paper, we show that quasi-alignments can be used to discover highly similar segments across multiple sequences from related or different genomes efficiently and accurately. Experiments on a large number of unaligned 16S rRNA sequences obtained from the Greengenes database show that the method is able to identify conserved regions which agree with known hypervariable regions in 16S rRNA. Furthermore, the experiments show that the proposed method scales well for large data sets with a run time that grows only linearly with the number and length of sequences, whereas for existing multiple sequence alignment heuristics the run time grows super-linearly. CONCLUSION: Quasi-alignment-based algorithms can detect highly similar regions and conserved areas across multiple sequences. Since the run time is linear and the sequences are converted into a compact clustering model, we are able to identify conserved regions fast or even interactively using a standard PC. Our method has many potential applications such as finding characteristic signature sequences for families of organisms and studying conserved and variable regions in, for example, 16S rRNA.

Enhancing our understanding of cellular processes will ultimately lead to the development of better therapeutic strategies. Completing incomplete biological pathways through utilizing probabilistic protein-protein interaction (PPI) networks is one approach towards establishing knowledge about cellular mechanisms. The existing complex/pathway membership methods are focused on uncovering candidate protein members from probabilistic PPI networks. In our previous work, we defined the pathway completion problem and developed a method that utilizes network motifs to complete incomplete biological pathways. Network motifs allow us to take into consideration the intrinsic local structures of pathways and identify the possible points of insertion of candidate proteins. However, our previous approach requires a complete and correct PPI network. In this paper, we extend our approach and use random walks on graphs to address the pathway completion problem with incomplete PPI networks. We evaluated our proposed method using three yeast probabilistic PPI networks and two yeast pathways from KEGG (Kyoto Encyclopedia of Genes and Genomes). Moreover, we compared the accuracy of our network motifs approach for pathway completion to the existing approach for pathway membership, which also utilizes random walks. Our experiments show that our new approach achieves similar or better accuracy. In addition, our method identifies the possible locations and connections of the candidate proteins in the incomplete pathway, which allows for more efficient experimental verification.

This paper describes one of the first attempts to model the temporal structure of massive data streams in real-time using data stream clustering. Recently, many data stream clustering algorithms have been developed which efficiently find a partition of the data points in a data stream. However, these algorithms disregard the information represented by the temporal order of the data points in the stream which for many applications is an important part of the data stream. In this paper we propose a new framework called Temporal Relationships Among Clusters for Data Streams (TRACDS) which allows us to learn the temporal structure while clustering a data stream. We identify, organize and describe the clustering operations which are used by state-of-the-art data stream clustering algorithms. Then we show that by defining a set of new operations to transform Markov Chains with states representing clusters dynamically, we can efficiently capture temporal ordering information. This framework allows us to preserve temporal relationships among clusters for any state-of-the-art data stream clustering algorithm with only minimal overhead. To investigate the usefulness of TRACDS, we evaluate the improvement of TRACDS over pure data stream clustering for anomaly detection using several synthetic and real-world data sets. The experiments show that TRACDS is able to considerably improve the results even if we introduce a high rate of incorrect time stamps which is typical for real-world data streams.

Phylogenetic tree analysis using molecular sequences continues to expand beyond the 16S rRNA marker. By addressing the multi-copy issue known as the intra-heterogeneity, this paper restores the focus in using the 16S rRNA marker. Through use of a novel learning and model building algorithm, the multiple gene copies are integrated into a compact complex signature using the Extensible Markov Model (EMM). The method clusters related sequence segments while preserving their inherent order to create an EMM signature for a microbial organism. A library of EMM signatures is generated from which samples are drawn for phylogenetic analysis. By matching the components of two signatures, referred to as quasi-alignment, the differences are highlighted and scored. Scoring quasi-alignments is done using adapted Karlin-Altschul statistics to compute a novel distance metric. The metric satisfies conditions of identity, symmetry, triangular inequality and the four point rule required for a valid evolution distance metric. The resulting distance matrix is input to PHYologeny Inference Package (PHYLIP) to generate phylogenies using neighbor joining algorithms. Through control of clustering in signature creation, the diversity of similar organisms and their placement in the phylogeny is explained. The experiments include analysis of genus Burkholderia, a random microbial sample spanning several phyla and a diverse sample that includes RNA of Eukaryotic origin. The NCBI sequence data for 16S rRNA is used for validation.

MOTIVATION: As next generation sequencing is rapidly adding new genomes, their correct placement in the taxonomy needs verification. However, the current methods for confirming classification of a taxon or suggesting revision for a potential misplacement relies on computationally intense multi-sequence alignment followed by an iterative adjustment of the distance matrix. Due to intra-heterogeneity issues with the 16S rRNA marker, no classifier is available for sub-genus level, which could readily suggest a classification for a novel 16S rRNA sequence. Metagenomics further complicates the issue by generating fragmented 16S rRNA sequences. This article proposes a novel alignment-free method for representing the microbial profiles using extensible Markov models (EMMs) with an extended Karlin-Altschul statistical framework similar to the classic alignment paradigm. We propose a log odds (LODs) score classifier based on Gumbel difference distribution that confirms correct classifications with statistical significance qualifications and suggests revisions where necessary. RESULTS: We tested our method by generating a sub-genus level classifier with which we re-evaluated classifications of 676 microbial organisms using the NCBI FTP database for the 16S rRNA. The results confirm current classification for all genera while ascertaining significance at 95%. Furthermore, this novel classifier isolates heterogeneity issues to a mere 12 strains while confirming classifications with significance qualification for the remaining 98%. The models require less memory than that needed by multi-sequence alignments and have better time complexity than the current methods. The classifier operates at sub-genus level, and thus outperforms the naive Bayes classifier of the RNA Database Project where much of the taxonomic analysis is available online. Finally, using information redundancy in model building, we show that the method applies to metagenomic fragment classification of 19 Escherichia coli strains. AVAILABILITY AND IMPLEMENTATION: Source code and binaries freely available for download at http://lyle.smu.edu/IDA/EMMSA/, implemented in JAVA and supported on MS Windows.