Biology subjects

Sakakibara, Y.

Publications and source records attributed to Sakakibara, Y..

At least 19 recordsLinked to original sources

bioRxiv2026

RIFT-VAE: grammar-conditioned pretraining and latent-space optimization for RNA inverse folding

BackgroundRNA inverse folding designs nucleotide sequences expected to adopt a prescribed secondary structure. Search-based solvers can optimize folding-model objectives effectively, but difficult targets can require extensive sampling, and structural optimization alone does not explicitly preserve the sequence distributions or conserved motifs of natural RNA families. MethodsWe developed RIFT-VAE, a Transformer-based conditional variational autoencoder that receives a context-free grammar parse-tree representation of a target secondary structure and generates nucleotide labels on the corresponding tree. The framework combines progressively richer grammar rules, self-refinement learning from generated structure-sequence pairs, and cross-entropy-method optimization in the learned latent space. We evaluated RNAfold minimum-free-energy agreement on an RNAcentral-derived test set and the EteRNA100 benchmark, compared the method with four search-based solvers under matched total time budgets, and examined GC-content control and covariance-model family annotation. ResultsThe complete pipeline achieved RNAfold-Correct/RNAfold-MCC values of 0.833/0.994 on the RNAcentral-derived test set and 0.760/0.977 on EteRNA100. Latent-space optimization accounted for the largest increase in exact structural recovery. Under a 3,600-s total budget on EteRNA100, sequences generated by RIFT-VAE improved the exact-match rate of every tested downstream search method when used as warm starts; the largest change was observed for RNAInverse (Correct, 0.297 to 0.803; MCC, 0.505 to 0.985). The pretrained model also produced sequences with measurable correct-family covariance-model hits and supported explicit GC-content conditioning. ConclusionsRIFT-VAE is best interpreted as a hybrid generative-search framework: pretraining supplies a structure- and family-informed proposal distribution, whereas latent optimization concentrates evaluations in high-scoring regions. The reported structural scores are specific to RNAfold minimum-free-energy validation and do not establish biochemical function. Orthogonal folding predictors, stricter homology-controlled splits, diversity-aware evaluation, architecture-matched dot-bracket ablations, and experimental assays remain priorities for validation.

bioinformatics↗

bioRxiv2026

A post-inflammatory C3-high astrocyte state persists after inflammatory stimulus withdrawal and is attenuated by JAK inhibition

Reactive astrocytes contribute to neuroinflammation and synaptic dysfunction, but it remains unclear whether transient inflammatory stimulation causes a persistent reactive state after the initial inflammatory stimulus is removed. Here, we investigated whether transient exposure to a defined inflammatory cytokine/complement cocktail induces a persistent reactive astrocyte state and examined the signaling mechanism underlying its maintenance. Human astrocytes were exposed to the inflammatory stimulus and subsequently subjected to stimulus washout, followed by time-course analyses to compare the reversibility of inflammatory gene expression after stimulus removal. Following washout, the expression of several inflammatory response genes, including CXCL10 and NF-{kappa}B-associated genes such as NFKBIA, TNFAIP3, and RELB, returned toward baseline levels. In contrast, C3 expression remained elevated, indicating persistence of a post-inflammatory C3-high astrocyte state after withdrawal of the inflammatory stimulus. Pharmacological inhibition of JAK signaling reduced persistent C3 expression to near-baseline levels, supporting the involvement of JAK-dependent signaling in maintenance of this persistent state. Together, these findings suggest that transient inflammatory stimulation induces a post-inflammatory persistent C3-high astrocyte state that is maintained even after broader inflammatory gene responses have subsided. This persistent C3-high component is pharmacologically attenuated by JAK inhibition, identifying JAK-dependent pathways as modulators of persistent astrocyte inflammatory reactivity.

neuroscience↗

bioRxiv2025

Genomic resources of Ascidiella aspersa and comparative analysis across tunicates reveal conserved class-level features and evolutionary diversification

BackgroundAscidiella aspersa is an invasive tunicate and one of the closest relatives of vertebrates. Despite its ecological nuisance, A. aspersa is anticipated to be a valuable model organism for developmental studies due to its remarkably transparent embryos. However, annotated genome assemblies and transcriptomic resources have not yet been fully established. Although several tunicate genomes have been sequenced, most lack annotations, and a comprehensive analysis across tunicates has not yet been conducted. ResultsWe performed de novo genome assembly and transcriptome analysis of A. aspersa, producing a high-quality 306.5 Mb genome assembly. The transcriptome was derived from nine different organs of adults and embryos at six developmental stages. Ab initio and homology-based gene predictions identified 24,504 genes with a BUSCO score of 92.2%. Functional annotation was added for 18,636 genes in the model. To understand the relative features of this species among tunicates, we conducted genome-wide comparative analysis using publicly available data from 35 other tunicate genomes across three classes, five orders, and 12 families, thus constructing gene models for 27 species with BUSCO scores >80%. Overall, phylogenetic analysis revealed a new hypothesis regarding the relationships among Phlebobranchia and Aplousobranchia families. Gene duplication analysis showed distinct contractions of gene families in some taxa with losses of specific DNA repair related genes that were shared among Thaliacea--these may have contributed to their evolutionary diversification. Tunicate genomes exhibited a high level of variation in genomic GC content (28.0%-42.7%). A. aspersa has the highest GC content in coding regions and the third position of codons among tunicate species, with changes in codon usage bias differing from other Ascidiidae species. We also constructed an online comparative tunicate genome database (TUNOME), that provides functional annotations of gene models and ortholog analyses based on these genomic and transcriptomic data. ConclusionsWe constructed genomic resources for A. aspersa and another 35 tunicate gene models. Comparative analysis reveals a variety in tunicate species genomes and characterizes class-level features. Our resources are expected to be a foundation for experimental studies involving non-model tunicates and for comparative analysis among tunicate species.

genomics↗

bioRxiv2025

Developmental Reorganization of Whole-Body Muscle Synergies During Overarm Throwing in Children

Overarm throwing is a uniquely human skill that requires precise whole-body coordination. Although throwing behavior emerges early in childhood, the neuromuscular mechanisms that support its development remain poorly characterized. Here, we provide novel evidence for the developmental reorganization of whole-body muscle synergies during maximum-effort throwing in preschool-aged (PS) and school-aged (SA) children. Electromyography was recorded from 16 muscles, and non-negative matrix factorization was applied to extract low-dimensional coordination modules (muscle synergies). We compared ball speed, number of synergies, synergy structure, and temporal consistency between groups. Ball speed was significantly higher in SA than PS (33.6 {+/-} 10.2 vs. 21.4 {+/-} 6.2 km/h, p < 0.05), reflecting improved performance. Yet, the number of synergies did not differ (PS: 6.0 {+/-} 1.1; SA: 6.4 {+/-} 1.3, p > 0.05), suggesting that the dimensionality of coordination is largely established by the preschool years. Instead, developmental improvements were driven by structural and temporal reorganization: trunk- and upper-limb synergies merged into a single module in SA, reflecting improved postural integration, while a bilateral soleus-dominant synergy fractionated into lateralized modules, reflecting increased lower-limb specialization. Moreover, the temporal variability of synergy activation was reduced in SA (p < 0.01), indicating that movement sequences became more precise and stable with development. These findings reveal that early gains in throwing arise not from expanding synergy number but from reorganizing their structure and sharpening temporal coordination, offering mechanistic insight into how complex whole-body skills are refined during childhood. Significance StatementThrowing is a hallmark of human motor behavior, requiring precise sequencing of whole-body muscle activity. Yet how children develop this ability has remained unclear. By applying muscle synergy analysis to electromyographic recordings of preschool and school-aged children performing maximum-effort throws, we found that improvements in performance were not due to an increase in synergy number but rather to structural reorganization and greater temporal precision. Specifically, trunk and upper-limb modules merged, lower-limb modules fractionated, and activation timing became more consistent. These results identify merging and fractionation as complementary mechanisms supporting developmental refinement of motor skills. More broadly, they provide a mechanistic framework for understanding how complex whole-body actions are acquired and offer markers for pediatric training and rehabilitation strategies.

neuroscience↗

bioRxiv2025

A groove brain-music interface for enhancing individual experience of urge to move

When we listen to music, we often feel a pleasurable urge to move to music, known as groove. While previous studies have identified musical features that contribute to the groove experience, such as syncopation and tempo, they also report individual differences in which kinds of music people experience groove. Therefore, recommending groove-eliciting music requires accounting for individual differences. In this study, we aimed to develop a groove brain-music interface (G-BMI) that generates personalized playlists to maximize each individuals groove experience, using a neurofeedback system based on in-ear EEG. Twenty-four participants listened to three high-groove and three low-groove musical excerpts and rated their "urge to move." Using these ratings and the recorded EEG, we trained two LASSO models to build the G-BMI. Model 1 predicted urge to move from acoustic features extracted with VGGish, a pretrained neural network. Model 2 classified EEG data as recorded during listening to high-groove or low-groove music. Using Model 1, we ranked 7,225 candidate songs by predicted groove and assembled one groove-augmenting and one groove-diminishing playlist. Using Models 1 and 2, we created two additional playlists that updated Model 1 and the ranking in real-time based on in-ear EEG. Participants then listened to all four playlists and rated them on items including "urge to move." The groove-augmenting playlist that incorporated EEG achieved the highest "urge to move" ratings. These findings suggest that a personalized neurofeedback system employing EEG can help maximize individual groove experience.

neuroscience↗

bioRxiv2025

Generative AI-driven artificial DNA design for enhancing inter-species gene activation and enzymatic degradation of PET

Conventional approaches to heterologous gene expression rely on codon optimization, which is limited to swapping synonymous codons and often fails to capture deeper adaptive changes. In contrast, naturally evolved orthologous genes between species often differ by more than just synonymous substitutions - they can include non-synonymous mutations, insertions, and deletions that confer functional adaptation to different host contexts. Here we present OrthologTransformer, a Transformer-based deep learning model that converts orthologous genes between species by learning from large-scale orthologous gene datasets curated for high-quality sequence alignments. The model recapitulates the full spectrum of evolutionary differences - from synonymous codon swaps to amino acid-changing mutations and indels - to predict a coding sequence optimized for a target species while preserving the proteins function. In extensive validation across diverse bacterial species pairs, OrthologTransformer significantly increased the conversion accuracy of generated genes to native target sequences compared to the original source genes, even for pairs with stark disparities in GC content and optimal growth temperature. The Transformers context-aware designs also favored conservative amino acid substitutions, maintaining protein functional integrity. As a proof-of-concept, an OrthologTransformer-designed PETase gene for Bacillus subtilis from the host species Ideonella sakaiensis was synthesized and expressed, yielding robust PET plastic-degradation activity that surpassed synonymous codon-optimized controls. These results establish OrthologTransformer as a powerful tool for de novo cross-species gene adaptation, transcending the limits of traditional codon optimization and enabling more effective heterologous gene performance in synthetic biology applications.

bioinformatics↗

bioRxiv2024

A chill brain-music interface for enhancing music chills with personalized playlists

Music chills are pleasurable experiences while listening to music, often accompanied by physical responses, such as goosebumps1,2. Enjoying music that induces chills is central to music appreciation, and engages the reward system in the brain3-5. However, the specific songs that trigger chills vary with individual preferences6, and the neural substrates associated with musical rewards differ among individuals7-9, making it challenging to establish a standard method for enhancing music chills. In this study, we developed the Chill Brain-Music Interface (C-BMI), a closed-loop neurofeedback system that uses in-ear electroencephalogram (EEG) for song selection. The C-BMI generates personalized playlists aimed at evoking chills by integrating individual song preferences and neural activity related to music reward processing. Twenty-four participants listened to both self-selected and other-selected songs, reporting higher pleasure levels and experiencing more chills in their self-selected songs. We constructed two LASSO regression models to support the C-BMI. Model 1 predicted pleasure based on the acoustic features of the self-selected songs. Model 2 classified the EEG responses when participants listened to self-selected versus other-selected songs. Model 1 was applied to over 7,000 candidate songs, predicting pleasure scores. We used these predicted scores and acoustic similarity to the self-selected songs to rank songs that were likely to induce pleasure. Using this ranking, four tailored playlists were generated. Two playlists were designed to augment pleasure by selecting top-ranked songs, one of which incorporated real-time pleasure estimates from Model 2 to continuously update Model 1 and refine song rankings. Additionally, two playlists aimed to diminish pleasure, with one updated using Model 2. We found that the pleasure-augmenting playlist with EEG-based updates elicited more chills and higher pleasure levels than pleasure-diminishing playlists. Our results indicate that C-BMI using in-ear EEG data can enhance music-induced chills.

neuroscience↗

bioRxiv2024

A Nostalgia Brain-Music Interface for Enhancing Nostalgia, Well-Being, and Memory Vividness in Young and Elderly Individuals

Music-evoked nostalgia has the potential to assist in recalling autobiographical memories and enhancing well-being. However, nostalgic music preferences vary from person to person, presenting challenges for applying nostalgia-based music interventions in clinical settings, such as a non-pharmacological approach. To address these individual differences, we developed the Nostalgia Brain-Music Interface (N-BMI), a neurofeedback system that recommends nostalgic songs tailored to each individual. This system is based on prediction models of nostalgic feelings, developed by integrating subjective nostalgia ratings, acoustic features and in-ear electroencephalographic (EEG) data during song listening. To test the effects of N-BMI on nostalgic feelings, well-being, and memory recall, seventeen elderly and seventeen young participants took part in the study. The N-BMI was personalized for each individual, and songs were recommended under two conditions: the "nostalgia condition", where songs were selected to enhance nostalgic feelings, and the "control condition", to reduce nostalgic feelings. We found nostalgic feelings, well-being, and memory vividness were significantly higher after listening to the recommended songs in the nostalgia condition compared to the control condition in both groups. This indicates that the N-BMI enhanced nostalgic feelings, well-being, and memory recall across both groups. The N-BMI paves the way for innovative therapeutic interventions, including non-pharmacological approaches.

neuroscience↗

bioRxiv2024

Transcriptomic dynamics of petal development in the one-day flower species, Japanese morning glory (Ipomoea nil)

Various aspects of Japanese morning glory (Ipomoea nil) petals, such as color, pattern, shape, flower opening time, and senescence, have been extensively studied. To facilitate such studies, transcriptome data were collected from flower petals at 3-h intervals over 3.5 days; the data was collected from 72 h before and 12 h after flower opening, accounting for 29 timepoints. Data analysis revealed substantial transcriptomic changes before and after flower opening. The expression patterns of cell division marker genes indicated that cell division practically stops at approximately 48 h before flower opening. Furthermore, the increased expression of genes encoding transporters for sugars, amino acids, nucleic acids, and autophagy-related genes was observed after flower opening, indicating the translocation of nutrients from senescing petal cells to other developing tissues. Correlations were found between the temporal expression patterns of the three transcriptional regulators and expression patterns of different sets of structural genes within the anthocyanin biosynthesis pathway, indicating differential reliance on each regulator for the activation of specific structural genes. Furthermore, clock genes were identified. Three copies of the clock gene ELF3 did not exhibit circadian rhythms, potentially allowing I. nil to adapt to high-latitude regions. The temporal transcriptome data and interactive database (https://ipomoeanil.nibb.ac.jp/fpkm/) offer valuable insights into gene expression dynamics, periodicity, and correlations and provide a crucial resource for further research on I. nil and other plant species.

plant biology↗

bioRxiv2023

Decoding host-microbiome interactions through co-expression network analysis within the non-human primate intestine

The gut microbiome affects the health status of the host through complex interactions with the hosts intestinal wall. These host-microbiome interactions may spatially vary along the physical and chemical environment of the intestine, but these changes remain unknown. This study investigated these intricate relationships through a gene co-expression network analysis based on dual transcriptome profiling of different intestinal sites--cecum, transverse colon, and rectum-- of the primate common marmoset. We proposed a gene module extraction algorithm based on graph theory to find tightly interacting gene modules of the host and the microbiome from a vast co-expression network. The 27 gene modules identified by this method, which include both host and microbiome genes, not only produced results consistent with previous studies regarding the host-microbiome relationships, but also provided new insights into microbiome genes acting as potential mediators in host-microbiome interplays. Specifically, we discovered associations between the host gene FBP1, a cancer marker, and polysaccharide degradation-related genes (pfkA and fucI) coded by Bacteroides vulgatus, as well as relationships between host B-cell specific genes (CD19, CD22, CD79B, PTPN6) and a tryptophan synthesis gene (trpB) coded by Parabacteroides distasonis. Furthermore, our proposed module extraction algorithm surpassed existing approaches by successfully defining more functionally related gene modules, providing insights for understanding the complex relationship between the host and the microbiome. IMPORTANCEWe unveiled the intricate dynamics of the host-microbiome interactions along the colon by identifying closely interacting gene modules from a vast gene co-expression network, constructed based on simultaneous profiling of both host and microbiome transcriptomes. Our proposed gene module extraction algorithm, designed to interpret inter-species interactions, enabled the identification of functionally related gene modules encompassing both host and microbiome genes, which was challenging with conventional modularity maximization algorithms. Through these identified gene modules, we discerned previously unrecognized bacterial genes that potentially mediate in known relationships between host genes and specific bacterial species. Our findings underscore the spatial variations in host-microbiome interactions along the colon, rather than displaying a uniform pattern throughout the colon.

microbiology↗

bioRxiv2023

Quantum algorithm for position weight matrix matching

We propose two quantum algorithms for a problem in bioinformatics, position weight matrix (PWM) matching, which aims to find segments (sequence motifs) in a biological sequence such as DNA and protein that have high scores defined by the PWM and are thus of informational importance related to biological function. The two proposed algorithms, the naive iteration method and the Monte-Carlo-based method, output matched segments, given the oracular accesses to the entries in the biological sequence and the PWM. The former uses quantum amplitude amplification (QAA) for sequence motif search, resulting in the query complexity scaling on the sequence length n, the sequence motif length m and the number of the PWMs K as [Formula], which means speedup over existing classical algorithms with respect to n and K. The latter also uses QAA, and further, quantum Monte Carlo integration for segment score calculation, instead of iteratively operating quantum circuits for arithmetic in the naive iteration method; then it provides the additional speedup with respect to m in some situation. As a drawback, these algorithms use quantum random access memories and their initialization takes O(n) time. Nevertheless, our algorithms keep the advantage especially when we search matches in a sequence for many PWMs in parallel.

bioinformatics↗

bioRxiv2021

Development of a metatranscriptomic analysis method for multiple intestinal sites and its application to the common marmoset intestine

BackgroundThe intestinal microbiome is closely related to host health, and metatranscriptomic analysis can assess the functional activity of microbiomes by quantifying the bacterial gene expression level, which helps to elucidate the interaction between the microbiome and the environment. However, functional changes in the microbiome along the host intestinal tract remain unknown, and previous analytical methods have limitations, such as potentially overlooking unknown genes due to dependence on existing databases and being unable to take full advantage of metatranscriptome to reveal the functional change among multiple environments. ResultTo close these gaps, we develop a novel method that integrates metagenome and metatranscriptome to analyze the functional activity of microbiomes between intestinal sites. This method reconstructs a reference metagenomic sequence across multiple intestinal sites, allowing the gene expression levels of microbiome including unknown bacterial genes to be compared between multiple sites. As a result of applying this method to metatranscriptomic analysis in the intestinal tract of common marmoset, the reconstructed metagenome covered most of the expressed genes and it revealed that the changes in bacterial gene expressions among the caecum, transverse colon, and faeces were more dynamic and sensitive to environmental shifts than its abundance. In typical, the coenzyme synthesis gene and antibacterial resistance gene were more highly expressed in the caecum and transverse colon than in faeces, while there was no significant change in abundance of these genes. ConclusionOur findings demonstrate that an analytical method that integrates metagenome and metatranscriptome in multiple intestinal sites captures functional changes in the microbiomes at the gene resolution level.

microbiology↗

bioRxiv2021

Informative RNA-base embedding for functional RNA structural alignment and clustering by deep representation learning

Effective embedding is being actively conducted by applying deep learning to biomolecular information. Obtaining better embedding enhances the quality of downstream analysis such as DNA sequence motif detection and protein function prediction. In this study, we adopt a pre-training algorithm for the effective embedding of RNA bases to acquire semantically rich representations, and apply it to two fundamental RNA sequence problems: structural alignment and clustering. By using the pre-learning algorithm to embed the four bases of RNA in a position-dependent manner using a large number of RNA sequences from various RNA families, a context-sensitive embedding representation is obtained. As a result, not only base information but also secondary structure and context information of RNA sequences are embedded for each base. We call this "informative base embedding" and use it to achieve accuracy superior to that of existing state-of-the-art methods in RNA structural alignment and RNA family clustering tasks. Furthermore, by performing RNA sequence alignment combining this informative base embedding with a simple Needleman-Wunsch alignment algorithm, we succeed in calculating a structural alignment in a time complexity O(n2) instead of the O(n6) time complexity of Sankoff-style algorithms.

bioinformatics↗

bioRxiv2021

Genomic style: yet another deep-learning approach to characterize bacterial genome sequences

MotivationBiological sequence classification is the most fundamental task in bioinformatics analysis. For example, in metagenome analysis, binning is a typical type of DNA sequence classification. In order to classify sequences, it is necessary to define sequence features. The k-mer frequency, base composition, and alignment-based metrics are commonly used. In contrast, in the field of image recognition using machine learning, image classification is broadly divided into those based on shape and those based on style. A style matrix was introduced as a method of expressing the style of an image (e.g., color usage and texture). ResultsWe propose a novel sequence feature, called genomic style, inspired by image classification approaches, for classifying and clustering DNA sequences. As with the style of images, the DNA sequence is considered to have a genomic style unique to the bacterial species, and the style matrix concept is applied to the DNA sequence. Our main aim is to introduce the genomics style as yet another basic sequence feature for metagenome binning problem in replace of the most commonly used sequence feature k-mer frequency. Performance evaluations show that our method using style matrix achieves the superior accuracy than state-of-the-art binning tools based on k-mer frequency.

bioinformatics↗

bioRxiv2021

Deep learning integration of molecular and interactome data for protein-compound interaction prediction

MotivationVirtual screening, which can computationally predict the presence or absence of protein-compound interactions, has attracted attention as a large-scale, low-cost, and short-term search method for seed compounds. Existing machine learning methods for predicting protein-compound interactions are largely divided into those based on molecular structure data and those based on network data. The former utilize information on proteins and compounds, such as amino acid sequences and chemical structures, while the latter utilize interaction network data, such as data on protein-protein interactions and compound-compound interactions. However, few attempts have been made to combine both types of data in molecular information and interaction networks. ResultsWe developed a deep learning-based method that integrates protein features, compound features, and multiple types of interactome data to predict protein-compound interactions. We designed three benchmark datasets with different difficulties and evaluated the performance on them. The performance evaluations show that our deep learning framework for integrating molecular structure data and interactome data outperforms state-of-the-art machine learning methods for protein-compound interaction prediction tasks. The performance improvement is proven to be statistically significant by the Wilcoxon signed-rank test. This reveals that the multi-interactome captures different perspectives than amino acid sequence homology and chemical structure similarity, and both type of data have a synergistic effect in improving prediction accuracy. Furthermore, experiments on three benchmark datasets show that our method is more robust than existing methods in accurately predicting interactions between proteins and compounds that are unseen in the training samples.

bioinformatics↗

bioRxiv2020

Chromosomal-scale De novo Genome Assemblies of Cynomolgus Macaque and Common Marmoset

Cynomolgus macaque (Macaca fascicularis) and common marmoset (Callithrix jacchus) have been widely used in human biomedical research. Their genomes were sequenced and assembled initially using short-read sequences, with the advent of massively parallel sequencing. However, the resulting contig sequences tended to remain fragmentary, and long-standing primate genome assemblies used the human genome as a reference for ordering and orienting the assembled fragments into chromosomes. Here we performed de novo genome assembly of these two species without any human genome-based bias observed in the genome assemblies released earlier. Firstly we assembled PacBio long reads, and the resultant contigs were scaffolded with Hi-C data. The scaffolded sequences obtained were further refined based on assembly results of alternate de novo assemblies and Hi-C contact maps by resolving identified inconsistencies. The final assemblies achieved N50 lengths of 149 Mb and 137 Mb for cynomolgus macaque and common marmoset, respectively, and the numbers of scaffolds longer than 10Mb are equal to their chromosome numbers. The high fidelity of our assembly is ascertained by concordance to the BAC-end read pairs observed for common marmoset, as well as a high resemblance of their karyotypic organization. Our assembly of cynomolgus macaque outperformed all the available assemblies of this species in terms of contiguity. The chromosome-scale genome assemblies produced in this study are valuable resources for non-human primate models and provide an important baseline in human biomedical research.

genomics↗

bioRxiv2020

Hamster PIWI proteins bind to piRNAs with stage-specific size variations during oocyte maturation

In animal gonads, transposable elements (TEs) are actively repressed to preserve genome integrity through the Piwi-interacting RNA (piRNA) pathway. In mice, piRNAs are most abundantly expressed in male germ cells, and form effector complexes with three distinct PIWI proteins. The depletion of individual Piwi genes causes male-specific sterility owing to severe defects in spermatogenesis with no discernible phenotype in female mice. Unlike mice, most other mammals have four PIWI genes, some of which are expressed in the ovary. Here, purification of PIWI complexes from oocytes of the golden hamster revealed that the size of the piRNAs loaded onto PIWIL1 changed during oocyte maturation. In contrast, PIWIL3, an ovary-specific PIWI in most mammals, associates with short piRNAs only in metaphase II oocytes, which coincides with intense phosphorylation of the protein. An improved high-quality genome assembly and annotation revealed that PIWIL1- and PIWIL3-associated piRNAs appear to share the 5'- ends of common piRNA precursors and are mostly derived from unannotated sequences with a diminished contribution from TE-derived sequences, most of which correspond to endogenous retroviruses (ERVs). Although binding sites for the transcription factor A-Myb are identified in the transcription start site regions of the testis piRNA clusters, the piRNA clusters in the ovary show no well-defined binding motifs in their upstream regions. These results show that hamster piRNA clusters are transcribed by different transcriptional factors in the ovary and testis, resulting in the generation of sex-specific piRNAs. Our findings show the complex and dynamic nature of biogenesis of piRNAs in hamster oocytes, and together with the new genome sequence generated, serve as the foundation for developing useful models to study the piRNA pathway in mammalian oocytes. Highlights- The size of PIWIL1-associated piRNAs changes during oocyte maturation - Phosphorylation of PIWIL3 in MII oocytes coincides with its association with small 19-nt piRNAs - Improved high-quality genome assembly and annotation identifies young endogenous retroviruses as major targets of piRNAs in hamster oocytes - PIWIL1- and PIWIL3-associated piRNAs share the 5'-ends of the common piRNA precursors in oocytes

molecular biology↗

bioRxiv2020

RNA secondary structure prediction using deep learning with thermodynamic integrations

RNA secondary structure prediction is one of the key technologies for revealing the essential roles of functional non-coding RNAs. Although machine learning-based rich-parametrized models have achieved extremely high performance in terms of prediction accuracy, the risk of overfitting for such models has been reported. In this work, we propose a new algorithm for predicting RNA secondary structures that uses deep learning with thermodynamic integration, thereby enabling robust predictions. Similar to our previous work, the folding scores, which are computed by a deep neural network, are integrated with traditional thermodynamic parameters to enable robust predictions. We also propose thermodynamic regularization for training our model without overfitting it to the training data. Our algorithm (MXfold2) achieved the most robust and accurate predictions in computational experiments designed for newly discovered non-coding RNAs, with significant 2-10 % improvements over our previous algorithm (MXfold) and standard algorithms for predicting RNA secondary structures in terms of F-value.

bioinformatics↗