Query Sequence Research Articles

Near-Duplicate Sequence Search at Scale for Large Language Model Memorization Evaluation

Recent studies show that large language models (LLM) unintendedly memorize part of the training data, which brings serious privacy risks. For example, it has been shown that over 1% of tokens generated unprompted by an LLM are part of sequences in the training data. However, current studies mainly focus on the exact memorization behaviors. In this paper, we propose to evaluate how many generated texts have near-duplicates (e.g., only differ by a couple of tokens out of 100) in the training corpus. A major challenge of conducting this evaluation is the huge computation cost incurred by near-duplicate sequence searches. This is because modern LLMs are trained on larger and larger corpora with up to 1 trillion tokens. What's worse is that the number of sequences in a text is quadratic to the text length. To address this issue, we develop an efficient and scalable near-duplicate sequence search algorithm in this paper. It can find (almost) all the near-duplicate sequences of the query sequence in a large corpus with guarantees. Specifically, the algorithm generates and groups the min-hash values of all the sequences with at least t tokens (as very short near-duplicates are often irrelevant noise) in the corpus in linear time to the corpus size. We formally prove that only 2 n+1/t+1 -1 min-hash values are generated for a text with n tokens in expectation. Thus the index time and size are reasonable. When a query arrives, we find all the sequences sharing enough min-hash values with the query using inverted indexes and prefix filtering. Extensive experiments on a few large real-world LLM training corpora show that our near-duplicate sequence search algorithm is efficient and scalable.

Extremely fast construction and querying of compacted and colored de Bruijn graphs with GGCAT.

Compacted de Bruijn graphs are one of the most fundamental data structures in computational genomics. Colored compacted de Bruijn graphs are a variant built on a collection of sequences and associate to each k-mer the sequences in which it appears. We present GGCAT, a tool for constructing both types of graphs, based on a new approach merging the k-mer counting step with the unitig construction step, as well as on numerous practical optimizations. For compacted de Bruijn graph construction, GGCAT achieves speed-ups of 3× to 21× compared with the state-of-the-art tool Cuttlefish 2. When constructing the colored variant, GGCAT achieves speed-ups of 5× to 39× compared with the state-of-the-art tool BiFrost. Additionally, GGCAT is up to 480× faster than BiFrost for batch sequence queries on colored graphs.

Pest Alert Tool-a web-based application for flagging species of concern in metabarcoding datasets.

Advances in high-throughput sequencing (HTS) technologies and their increasing affordability have fueled environmental DNA (eDNA) metabarcoding data generation from freshwater, marine and terrestrial ecosystems. Research institutions worldwide progressively employ HTS for biodiversity assessments, new species discovery and ecological trendmonitoring. Moreover, even non-scientists can now collect an eDNA sample, send it to a specialized laboratory for analysis and receive in-depth biodiversity record from a sampling site. This offers unprecedented opportunities for biodiversity assessments across wide temporal and spatial scales. The large volume of data produced by metabarcoding also enables incidental detection of species of concern, including non-indigenous and pathogenic organisms. We introduce an online app-Pest Alert Tool-for screening nuclear small subunit 18S ribosomal RNA and mitochondrial cytochrome oxidase subunit I datasets for marine non-indigenous species as well as unwanted and notifiable marine organisms in New Zealand. The output can be filtered by minimum length of the query sequence and identity match. For putative matches, a phylogenetic tree can be generated through the National Center for Biotechnology Information's BLAST Tree View tool, allowing for additional verification of the species of concern detection. The Pest Alert Tool is publicly available at https://pest-alert-tool-prod.azurewebsites.net/.

A Confusion Method for the Protection of User Topic Privacy in Chinese Keyword-based Book Retrieval

In this article, aiming at a Chinese keyword-based book search service, from a technological perspective, we propose to modify a user query sequence carefully to confuse the user query topics and thus protect the user topic privacy on the untrusted server, without compromising the accuracy of each book search service. First, we propose a client-based framework for the privacy protection of book search, and then a privacy model to formulate the constraints in terms of accuracy, efficiency, and security, which the cover queries generated based on a user query sequence should meet. Second, we present a modification algorithm for a user query sequence, based on some heuristic strategies, which can quickly generate a cover query sequence meeting the privacy model by replacing, deleting, and adding keywords for each user query. Finally, both theoretical analysis and experimental evaluation demonstrate the effectiveness of the proposed approach, i.e., which can improve the security of users’ topic privacy on the untrusted server without compromising the efficiency, accuracy, and usability of an existing Chinese keyword book search service, so it has a positive impact for the construction of a privacy-preserving text retrieval platform under an untrusted network environment.

Comparative genomic analyses of the clinically-derived Winkia strain NY0527: the reassignment of W. neuii subsp. neuii and W. neuii subsp. antitratus into two separate species and insights into their virulence characteristics.

Winkia neuii, previously known as Actinomyces neuii, is increasingly recognized as a causative agent of various human infections, while its taxonomy and genomic insights are still understudied. A Winkia strain NY0527 was isolated from the hip abscess of a patient, and its antibiotic susceptibility was assessed. The genome was hybrid assembled from long-reads and short-reads sequencing. Whole-genome-based analyses on taxa assignment, strain diversity, and pathogenesis were conducted. The strain was found to be highly susceptible to beta-lactam antibiotics, but resistant to erythromycin, clindamycin, and amikacin. The complete genome sequences of this strain were assembled and found to consist of a circular chromosome and a circular plasmid. Sequence alignment to the NCBI-nt database revealed that the plasmid had high sequence identity (>90%) to four Corynebacterium plasmids, with 40-50% query sequence coverage. Furthermore, the plasmid was discovered to possibly originate from the sequence recombination events of two Corynebacterium plasmid families. Phylogenomic tree and genomic average nucleotide identity analyses indicated that many Winkia sp. strains were still erroneously assigned as Actinomyces sp. strains, and the documented subspecies within W. neuii should be reclassified as two separate species (i.e., W. neuii and W. anitratus). The core genome of each species carried a chromosome-coded beta-lactamase expression repressor gene, which may account for their broadly observed susceptibility to beta-lactam antibiotics in clinical settings. Additionally, an ermX gene that expresses fluoroquinolone resistance was shared by some W. neuii and W. anitratus strains, possibly acquired by IS6 transposase-directed gene transfer events. In contrast, tetracycline resistance genes were exclusively carried by W. neuii strains. In particular, W. neuii was found to be more pathogenic than W. anitratus by encoding more virulence factors (i.e., 35-38 in W. neuii vs 27-31 in W. anitratus). Moreover, both species encoded two core pathogenic virulence factors, namely hemolysin and sialidase, which may facilitate their infections by expressing poreformation, adhesion, and immunoglobulin deglycosylation activities. This study highlights the underappreciated taxonomic diversity of Winkia spp. and provides populational genomic insights into their antibiotic susceptibility and pathogenesis for the first time, which could be helpful in the clinical diagnosis and treatment of Winkia spp. infections.

Deterministic Model of Information Queries Batch Processing in a Multithreaded Server

The analytical model of processing a batch of similar requests in a multithreaded server is considered. The model is based on the high determinism of the duration of individual stages of request processing and the moments of events leading to the switching of the processor from one processing program thread to another. Unlike other deterministic models based on the theory of schedules, in this case the goal is not to find the optimal distribution of work across processors, but to study the dependence of the main characteristics of the software server (packet processing time, marginal performance) on the characteristics of the request and the size of the batch. The model concerns the situa­tion when distributed system with intensive flows of information queries is developed and special measures to ensure the required performance must be taken. One of these measures is the introduction of batch query processing tools, allowing an array of similar information queries to be processed in a single service call. The introduction of such tools usually allows you to repeatedly increase the rate of processing queries in the system and very often becomes a real «lifesaver» for customers who issue, for example, hundreds of thousands of queries per day or more. It is quite likely that a sequence of ordinary single queries cannot provide a solution, and batch processing simply has no alternative

ACP-MLC: A two-level prediction engine for identification of anticancer peptides and multi-label classification of their functional types

Anticancer peptides (ACPs), a series of short bioactive peptides, are promising candidates in fighting against cancer due to their high activity, low toxicity, and not likely cause drug resistance. The accurate identification of ACPs and classification of their functional types is of great importance for investigating their mechanisms of action and developing peptide-based anticancer therapies. Here, we provided a computational tool, called ACP-MLC, to address binary classification and multi-label classification of ACPs for a given peptide sequence. Briefly, ACP-MLC is a two-level prediction engine, in which the 1st-level model predicts whether a query sequence is an ACP or not by random forest algorithm, and the 2nd-level model predicts which tissue types the sequence might target by the binary relevance algorithm. Development and evaluation by high-quality datasets, our ACP-MLC yielded an area under the receiver operating characteristic curve (AUC) of 0.888 on the independent test set for the 1st-level prediction, and obtained 0.157 hamming loss, 0.577 subset accuracy, 0.802 F1-scoremacro, and 0.826 F1-scoremicro on the independent test set for the 2nd-level prediction. A systematic comparison demonstrated that ACP-MLC outperformed existing binary classifiers and other multi-label learning classifiers for ACP prediction. Finally, we interpreted the important features of ACP-MLC by the SHAP method. User-friendly software and the datasets are available at https://github.com/Nicole-DH/ACP-MLC. We believe that the ACP-MLC would be a powerful tool in ACP discovery.

Genomic Analysis of Amphioxus Reveals a Wide Range of Fragments Homologous to Viral Sequences.

Amphioxus species are considered living fossils and are important in the evolutionary study of chordates and vertebrates. To explore viral homologous sequences, a high-quality annotated genome of the Beihai amphioxus (Branchiostoma belcheri beihai) was examined using virus sequence queries. In this study, 347 homologous fragments (HFs) of viruses were identified in the genome of B. belcheri beihai, of which most were observed on 21 genome assembly scaffolds. HFs were preferentially located within protein-coding genes, particularly in their CDS regions and promoters. A range of amphioxus genes with a high frequency of HFs is proposed, including histone-related genes that are homologous to the Histone or Histone H2B domains of viruses. Together, this comprehensive analysis of viral HFs provides insights into the neglected role of viral integration in the evolution of amphioxus.

ElasticBLAST: accelerating sequence search via cloud computing

BackgroundBiomedical researchers use alignments produced by BLAST (Basic Local Alignment Search Tool) to categorize their query sequences. Producing such alignments is an essential bioinformatics task that is well suited for the cloud. The cloud can perform many calculations quickly as well as store and access large volumes of data. Bioinformaticians can also use it to collaborate with other researchers, sharing their results, datasets and even their pipelines on a common platform.ResultsWe present ElasticBLAST, a cloud native application to perform BLAST alignments in the cloud. ElasticBLAST can handle anywhere from a few to many thousands of queries and run the searches on thousands of virtual CPUs (if desired), deleting resources when it is done. It uses cloud native tools for orchestration and can request discounted instances, lowering cloud costs for users. It is supported on Amazon Web Services and Google Cloud Platform. It can search BLAST databases that are user provided or from the National Center for Biotechnology Information.ConclusionWe show that ElasticBLAST is a useful application that can efficiently perform BLAST searches for the user in the cloud, demonstrating that with two examples. At the same time, it hides much of the complexity of working in the cloud, lowering the threshold to move work to the cloud.

Non-Autoregressive Sparse Transformer Networks for Pedestrian Trajectory Prediction

Pedestrian trajectory prediction is an important task in practical applications such as automatic driving and surveillance systems. It is challenging to effectively model social interactions among pedestrians and capture temporal dependencies. Previous methods typically emphasized social interactions among pedestrians but ignored the temporal consistency of predictions and suffered from superfluous interactions by dense undirected graphs, resulting in a considerable deviance from reality. In addition, autoregressive approaches predicted future locations conditioning on previous predictions one by one, which would lead to error accumulation and time consuming. To address these issues, we present Non-autoregressive Sparse Transformer (NaST) networks for pedestrian trajectory prediction. Specifically, NaST models sparse spatial interactions and sparse temporal dependency via a sparse spatial transformer and a sparse temporal transformer separately. Different from previous predictions such as RNN-based approaches, the transformer decoder works in non-autoregressive pattern and predicts all the future locations at one time from a query sequence, which could avoid the error accumulation and be less computationally intensive. We evaluate our proposed method on the ETH and UCY datasets, and the experimental results show our method outperforms comparative state-of-the-art methods.

SCAMPP: Scaling Alignment-Based Phylogenetic Placement to Large Trees.

Phylogenetic placement, the problem of placing a "query" sequence into a precomputed phylogenetic "backbone" tree, is useful for constructing large trees, performing taxon identification of newly obtained sequences, and other applications. The most accurate current methods, such as pplacer and EPA-ng, are based on maximum likelihood and require that the query sequence be provided within a multiple sequence alignment that includes the leaf sequences in the backbone tree. This approach enables high accuracy but also makes these likelihood-based methods computationally intensive on large backbone trees, and can even lead to them failing when the backbone trees are very large (e.g., having 50,000 or more leaves). We present SCAMPP (SCaling AlignMent-based Phylogenetic Placement), a technique to extend the scalability of these likelihood-based placement methods to ultra-large backbone trees. We show that pplacer-SCAMPP and EPA-ng-SCAMPP both scale well to ultra-large backbone trees (even up to 200,000 leaves), with accuracy that improves on APPLES and APPLES-2, two recently developed fast phylogenetic placement methods that scale to ultra-large datasets. EPA-ng-SCAMPP and pplacer-SCAMPP are available at https://github.com/chry04/PLUSplacer.

Isolation and Identification of Xylella fastidiosa that Cause Oleander Leaf Scorch

During spring of 2022, leaf scorchsymptom was observed in Nerium oleander plant in Erbil city. Surveys were carried out in three main Erbil streets (60-meter, 120 meter and Qasmlo) to determine the occurrence and distribution of oleander leaf scorch. Oleander leaf scorch (OLS) is a disease caused by xylem limited bacterium Xylella fastidiosa. The percentage of plants with Oleander leaf scorch infection ranged usually between 23-91%, and the disease was assessed the highest disease rate based on disease rating scale was recorded in 60-meter street is 91.95%, Qasmlo street is 89.07% and the lowest disease rate was observed in 120-meter street is 23.17%. In this study the bacterium was detected after isolated in laboratory as Gram stain test was performed and the isolated bacterium was Gram negative, rod shaped, non-flagellate and non-motile also identified by PCR were performed with two primers forward primer 16S (AGAG TTTG ATCC TGGC TCAG) and reverse primer 16S (GGCT ACCT TGTT ACGA CTT), amplified a 1200 bp fragment from the isolated OLS. The Analysis of BLAST revealed that the highest identity number query sequences were (100%) identified and submitted in GenBank and have taken accession number Xylella fastidiosa subsp. Sandyi (ON131099) and Xylella fastidiosa subsp. Sandyi (ON131100) also the phylogenetic results show the each identified bacteria are grouped in one clade with other sequences were stored in NCBI GenBank. This is a first report of a ‘Xylella fastidiosa.in Erbil city.

ProteInfer, deep neural networks for protein functional inference.

Predicting the function of a protein from its amino acid sequence is a long-standing challenge in bioinformatics. Traditional approaches use sequence alignment to compare a query sequence either to thousands of models of protein families or to large databases of individual protein sequences. Here we introduce ProteInfer, which instead employs deep convolutional neural networks to directly predict a variety of protein functions - Enzyme Commission (EC) numbers and Gene Ontology (GO) terms - directly from an unaligned amino acid sequence. This approach provides precise predictions which complement alignment-based methods, and the computational efficiency of a single neural network permits novel and lightweight software interfaces, which we demonstrate with an in-browser graphical interface for protein function prediction in which all computation is performed on the user's personal computer with no data uploaded to remote servers. Moreover, these models place full-length amino acid sequences into a generalised functional space, facilitating downstream analysis and interpretation. To read the interactive version of this paper, please visit https://google-research.github.io/proteinfer/.

Intragenomic rearrangements involving 5′-untranslated region segments in SARS-CoV-2, other betacoronaviruses, and alphacoronaviruses

BackgroundVariation of the betacoronavirus SARS-CoV-2 has been the bane of COVID-19 control. Documented variation includes point mutations, deletions, insertions, and recombination among closely or distantly related coronaviruses. Here, we describe yet another aspect of genome variation by beta- and alphacoronaviruses that was first documented in an infectious isolate of the betacoronavirus SARS-CoV-2, obtained from 3 patients in Hong Kong that had a 5′-untranslated region segment at the end of the ORF6 gene that in its new location translated into an ORF6 protein with a predicted modified carboxyl terminus. While comparing the amino acid sequences of translated ORF8 genes in the GenBank database, we found a subsegment of the same 5′-UTR-derived amino acid sequence modifying the distal end of ORF8 of an isolate from the United States and decided to carry out a systematic search.MethodsUsing the nucleotide and in the case of SARS-CoV-2 also the translated amino acid sequence in three reading frames of the genomic termini of coronaviruses as query sequences, we searched for 5′-UTR sequences in regions other than the 5′-UTR in SARS-CoV-2 and reference strains of alpha-, beta-, gamma-, and delta-coronaviruses.ResultsWe here report numerous genomic insertions of 5′-untranslated region sequences into coding regions of SARS-CoV-2, other betacoronaviruses, and alphacoronaviruses, but not delta- or gammacoronaviruses. To our knowledge this is the first systematic description of such insertions. In many cases, these insertions would change viral protein sequences and further foster genomic flexibility and viral adaptability through insertion of transcription regulatory sequences in novel positions within the genome. Among human Embecorivus betacoronaviruses, for instance, from 65% to all of the surveyed sequences in publicly available databases contain inserted 5′-UTR sequences.ConclusionThe intragenomic rearrangements involving 5′-untranslated region sequences described here, which in several cases affect highly conserved genes with a low propensity for recombination, may underlie the generation of variants homotypic with those of concern or interest and with potentially differing pathogenic profiles. Intragenomic rearrangements thus add to our appreciation of how variants of SARS-CoV-2 and other beta- and alphacoronaviruses may arise.

Monostephanostomum mesospinosum (Digenea: Acanthocolpidae) Infecting the Coastal Trevally Carangoides caeruleopinnatus (Carangiformes: Carangidae) from the Marine Water off Saudi Arabia

Background: Carangids, like any other fish, can become infected with a variety of external and internal parasites. Little information on parasite infestations in these fish species is known. This study aimed to provide light on parasites that infect caragnids in Saudi Arabia. Methods: A total of 20 Carangoides caeruleopinnatus specimens were collected from the Red Sea (Saudi Arabia) and then examined for the presence of digenetic trematodes. Parasitic species were isolated and studied morphologically by using light microscopic examination and molecularly via the partial sequencing of 18S rRNA and 28S rRNA genes. Result: Only one parasitic species has been identified. This parasite is morphologically and morphometric compatible with previously Monostephanostomum mesospinosum, identified from Carangoides malabaricus in VSK, BOB. Phylogeny revealed that these putative acanthocolpid species nested well within a clade clustering Monostephanostomum species, which along with morphological data, suggests it is a member of the genus Monostephanostomum. Query sequences showed identities of 97.99% and 97.98% for 18S (EF506762.1.1) and 28S rRNA (EF506763.1) of Monostephanostomum nolani, respectively. This study reflects the third account of this genus as endoparasitic taxa of various carangid fish, as well as providing novel DNA data for this species.

A Guide to Gene-Centric Analysis Using TreeSAPP.

Gene-centric analysis is commonly used to chart the structure, function, and activity of microbial communities in natural and engineered environments. A common approach is to create custom ad hoc reference marker gene sets, but these come with the typical disadvantages of inaccuracy and limited utility beyond assigning query sequences taxonomic labels. The Tree-based Sensitive and Accurate Phylogenetic Profiler (TreeSAPP) software package standardizes analysis of phylogenetic and functional marker genes and improves predictive performance using a classification algorithm that leverages information-rich reference packages consisting of a multiple sequence alignment, a profile hidden Markov model, taxonomic lineage information, and a phylogenetic tree. Here, we provide a set of protocols that link the various analysis modules in TreeSAPP into a coherent process that both informs and directs the user experience. This workflow, initiated from a collection of candidate reference sequences, progresses through construction and refinement of a reference package to marker identification and normalized relative abundance calculations for homologous sequences in metagenomic and metatranscriptomic datasets. The alpha subunit of methyl-coenzyme M reductase (McrA) involved in biological methane cycling is presented as a use case given its dual role as a phylogenetic and functional marker gene driving an ecologically relevant process. These protocols fill several gaps in prior TreeSAPP documentation and provide best practices for reference package construction and refinement, including manual curation steps from trusted sources in support of reproducible gene-centric analysis. © 2023 The Authors. Current Protocols published by Wiley Periodicals LLC. Basic Protocol 1: Creating reference packages Support Protocol 1: Installing TreeSAPP Support Protocol 2: Annotating traits within a phylogenetic context Basic Protocol 2: Updating reference packages Basic Protocol 3: Calculating relative abundance of genes in metagenomic and metatranscriptomic datasets.

Molecular characterization of a chalcone synthase gene RhCHS from Rhododendron × hybridum Hort

Chalcone synthase (CHS) plays a vital role in anthocyanin biosynthesis pathway, which is associated with petal color of flower. To date, lots of CHS genes have been obtained from plants, while few were from Rhododendron genus. In this study we got a new CHS gene named RhCHS (MW358095) from Rhododendron × hybridum Hort. It had a 2040 bp coding region consisting of two exons and one intron. By using the deduced RhCHS protein as a query sequence, 15 CHS homologous family genes with sequence similarity from 60% to 98%, designated as RgCHS-D(x), were retrieved from the genome assembly of Rhododendron griersonianum (RGv1.1) by TBlastN. 12 CHS family genes were found locating in No.9 chromosome arranged in clusters, while only 3 of them exhibited in No.1, 2, and 8 chromosomes, respectively. The results revealed gene duplication of CHS in evolutionary process. Multiple alignment of the deduced amino acid sequence of RhCHS showed high similarity of the active site, the catalytic residue, and the signature motif, the conserved characteristics of which were also exhibited in the tertiary structure prediction of the RhCHS, as well as the phylogenetic tree, all these demonstrated the RhCHS belonging to the type III PKS superfamily. HPLC-MS/MS of flower petals detected the total concentration of CC, DC, and PelC. These anthocyanidins showed an overall increasing trend during the flowering period and reached the peak in the full-blooming stage, which was consistence with the changeable rule of RhCHS expression level. The promoter, which was 1507 bp exhibiting high β-glucuronidase (GUS) staining activity, was predicted containing many cis-acting elements, especially light and transcription factor such as bHLH, MYB, WRKY, Dof, and ERF. In short, this study may provide the help to Rhododendron × hybridum Hort. not only in the mechanism research of petals color exhibition, but also in molecular breeding of CHS practice value.

SCAMPP+FastTree: improving scalability for likelihood-based phylogenetic placement.

Phylogenetic placement is the problem of placing 'query' sequences into an existing tree (called a 'backbone tree'). One of the most accurate phylogenetic placement methods to date is the maximum likelihood-based method pplacer, using RAxML to estimate numeric parameters on the backbone tree and then adding the given query sequence to the edge that maximizes the probability that the resulting tree generates the query sequence. Unfortunately, this way of running pplacer fails to return valid outputs on many moderately large backbone trees and so is limited to backbone trees with at most ∼10000 leaves. SCAMPP is a technique to enable pplacer to run on larger backbone trees, which operates by finding a small 'placement subtree' specific to each query sequence, within which the query sequence are placed using pplacer. That approach matched the scalability and accuracy of APPLES-2, the previous most scalable method. Here, we explore a different aspect of pplacer's strategy: the technique used to estimate numeric parameters on the backbone tree. We confirm anecdotal evidence that using FastTree instead of RAxML to estimate numeric parameters on the backbone tree enables pplacer to scale to much larger backbone trees, almost (but not quite) matching the scalability of APPLES-2 and pplacer-SCAMPP. We then evaluate the combination of these two techniques-SCAMPP and the use of FastTree. We show that this combined approach, pplacer-SCAMPP-FastTree, has the same scalability as APPLES-2, improves on the scalability of pplacer-FastTree and achieves better accuracy than the comparably scalable methods. https://github.com/gillichu/PLUSplacer-taxtastic. Supplementary data are available at Bioinformatics Advances online.

HMMerge: an ensemble method for multiple sequence alignment.

Despite advances in method development for multiple sequence alignment over the last several decades, the alignment of datasets exhibiting substantial sequence length heterogeneity, especially when the input sequences include very short sequences (either as a result of sequencing technologies or of large deletions during evolution) remains an inadequately solved problem. We present HMMerge, a method to compute an alignment of datasets exhibiting high sequence length heterogeneity, or to add short sequences into a given 'backbone' alignment. HMMerge builds on the technique from its predecessor alignment methods, UPP and WITCH, which build an ensemble of profile HMMs to represent the backbone alignment and add the remaining sequences into the backbone alignment using the ensemble. HMMerge differs from UPP and WITCH by building a new 'merged' HMM from the ensemble, and then using that merged HMM to align the query sequences. We show that HMMerge is competitive with WITCH, with an advantage over WITCH when adding very short sequences into backbone alignments. HMMerge is freely available at https://github.com/MinhyukPark/HMMerge. Supplementary data are available at Bioinformatics Advances online.

Problematic Unordered Queries in Temporal Moment Measurement by Using Natural Language

This study examines the difficulty in measuring temporal moments by using natural language (TMMNL) in the untrimmed video. The purpose of TMMNL is to use natural language query to find a specific moment within a lengthy video. It’s a challenging task since other, closely related activities may divert attention from the target temporal moment. This issue has been addressed by existing research employing computer vision techniques like reinforcement, anchor, and ranking. In this research, we not only propose a TMMNL solution and show how to use natural language query to find the required moment, but we also identify a novel issue: if the given natural language query is unordered (without proper subject, verb and object), the system will have trouble understanding and the network may perform poorly. Previous methods perform poorer when a query is unordered and cannot able to retrieve the relevant moment as a result overall performance degradation. We introduce novel the-visual, the-action, the-object, and the-connecting words concept to address the problem of unordered query in TMMNL. Graph Convolutions with Latent variable for Visual-Textual Network (GCL-VTN) is our suggested network, which has three components: (1) visual-graph-convolution (visual GC), (2) textual-graph-convolution (textual GC), and (3) compatible method for learning embedding’s (CMLE). Visual-nodes in the visual GC detect regional attributes, object, and actor information in the same way as textual-nodes in the textual GC maintain word sequence using grammar-based query rules. A compatible method for learning embedding (CMLE) is also proposed, which integrates different modalities (moment, query) and trained grammar-based words into the same embedding space. In order to align and keep the query sequence, we also incorporate a stochastic latent variable in the CMLE that has prior and posterior distributions. The posterior distribution deals with both visual-textual data and works when the query is in the correct sequence or based on grammar rules. The prior distribution only deals with textual data and is effective when the query is unordered or not based on grammar rules. TACoS, Charades-STA, and activityNet-Captions are state-of-the-art; our GCL-VTN exceeds them all.