Machine learning trial to detect sex differences in simple sticker arts of 1606 preschool children.

BACKGROUND: Previous studies suggested that drawings made by preschool boys and girls show distinguishable differences. However, children's drawings on their own are too complexly determined and inherently ambiguous to be a reliable indicator. In the present study, we attempted to develop a machine learning algorithm for classification of sex of the subjects using children's artworks. METHODS: We studied three types of simple sticker artworks from 1606 Japanese preschool children aged 51-83 months (803 boys and 803 girls). Those artworks were processed into digitalized data. Simulated data based on the original data were also generated. Logistic regression approach was applied to each dataset to make a classifier, and run on each dataset in a stratified ten-fold cross-validation with hyperparameter tuning. A probability score was calculated in each sample and utilized for sex classification. Prediction performance was evaluated using accuracy, recall, and precision scores, as well as learning curves. RESULTS: Two models created from the original and simulated data showed comparably low metrics. The distributions of probability scores in the samples from boys and girls mostly overlapped and were indistinguishable. Learning curves of the models showed an extremely under-fitted pattern. CONCLUSIONS: Our machine learning algorithm was unable to distinguish simple sticker arts created by boys and girls. More complex tasks will enable to develop an accurate classifier.

Automated urinary sediment detection for Fabry disease using deep-learning algorithms.

Fabry disease is a congenital lysosomal storage disease, and most of these cases develop organ damage in middle age. There are some promising therapeutic options for this disorder, which can stabilize the progression of the disease. However, a long delay in diagnosis prevents early intervention, resulting in treatment failure. Because Fabry disease is a rare disease, it is not well recognized and disease specific screening tests are rarely performed. Hence, a novel approach to for detecting patients with a widely practiced clinical test is crucial for the early detection of the disease. Recently, decision support systems based on artificial intelligence (AI) have been developed in many clinical fields. However, the construction of these models requires datasets from a large number of samples; this aspect is one of the main obstacles in AI-based approaches for rare diseases. In this study, with a novel image amplification method to construct the dataset for AI-model training, we built the deep neural-network model to detect Fabry cases from their urine samples. Sensitivity, specificity, and the AUC of the models on validation dataset were 0.902 (95% CI, 0.900-0.903), 0.977 (0.950-0.980), and 0.968 (0.964-0.972), respectively. This model could also extract disease-specific findings that are interpretable with human recognition. These results indicate that we can apply novel AI models for rare diseases based on this image amplification method we developed. We expect this approach could contribute to the diagnosis of Fabry disease. Synopsis: This is the first reported AI-based decision support system to detect undiagnosed Fabry cases, and our new image amplification method will contribute to the AI models for other rare disorders.

Collection of 2429 constrained headshots of 277 volunteers for deep learning.

Deep learning has rapidly been filtrating many aspects of human lives. In particular, image recognition by convolutional neural networks has inspired numerous studies in this area. Hardware and software technologies as well as large quantities of data have contributed to the drastic development of the field. However, the application of deep learning is often hindered by the need for big data and the laborious manual annotation thereof. To experience deep learning using the data compiled by us, we collected 2429 constrained headshot images of 277 volunteers. The collection of face photographs is challenging in terms of protecting personal information; we therefore established an online procedure in which both the informed consent and image data could be obtained. We did not collect personal information, but issued agreement numbers to deal with withdrawal requests. Gender and smile labels were manually and subjectively annotated only from the appearances, and final labels were determined by majority among our team members. Rotated, trimmed, resolution-reduced, decolorized, and matrix-formed data were allowed to be publicly released. Moreover, simplified feature vectors for data sciences were released. We performed gender and smile recognition by building convolutional neural networks based on the Inception V3 model with pre-trained ImageNet data to demonstrate the usefulness of our dataset.

Identification of epigenetic memory candidates associated with gestational age at birth through analysis of methylome and transcriptional data.

Preterm birth is known to be associated with chronic disease risk in adulthood whereby epigenetic memory may play a mechanistic role in disease susceptibility. Gestational age (GA) is the most important prognostic factor for preterm infants, and numerous DNA methylation alterations associated with GA have been revealed by epigenome-wide association studies. However, in human preterm infants, whether the methylation changes relate to transcription in the fetal state and persist after birth remains to be elucidated. Here, we identified 461 transcripts associated with GA (range 23-41 weeks) and 2093 candidate CpG sites for GA-involved epigenetic memory through analysis of methylome (110 cord blood and 47 postnatal blood) and transcriptional data (55 cord blood). Moreover, we discovered the trends of chromatin state, such as polycomb-binding, among these candidate sites. Fifty-four memory candidate sites showed correlation between methylation and transcription, and the representative corresponding gene was UCN, which encodes urocortin.

Diversification of CpG-Island Promoters Revealed by Comparative Analysis Between Human and Rhesus Monkey Genomes.

While CpG dinucleotides are significantly reduced compared to other dinucleotides in mammalian genomes, they can congregate and form CpG islands, which localize around the 5' regions of genes, where they function as promoters. CpG-island promoters are generally unmethylated and are often found in housekeeping genes. However, their nucleotide sequences and existence per se are not conserved between humans and mice, which may be due to evolutionary gain and loss of the regulatory regions. In this study, human and rhesus monkey genomes, with moderately conserved sequences, were compared at base resolution. Using transcription start site data, we first validated our methods' ability to identify orthologous promoters and indicated a limitation using the 5' end of curated gene models, such as NCBI RefSeq, as their transcription start sites. We found that, in addition to deamination mutations, insertions and deletions of bases, repeats, and long fragments contributed to the mutations of CpG dinucleotides. We also observed that the G + C contents tended to change in CpG-poor environments, while CpG content was altered in G + C-rich environments. While loss of CpG islands can be caused by gradual decreases in CpG sites, gain of these islands appear to require two distinct nucleotide altering steps. Taken together, our findings provide novel insights into the process of acquisition and diversification of CpG-island promoters in vertebrates.

Frequent retrotransposition of endogenous genes in ERCC2-deficient cells derived from a patient with xeroderma pigmentosum.

BACKGROUND: Retrotransposition of protein-coding genes is thought to occur due to the existence of numerous processed pseudogenes in both animals and plants. Unlike retrotransposons including Alu and LINE-1, direct evidence of such retrotransposition events has not been reported to date. Even if such an event occurs in a somatic cell, it is almost impossible to detect it using bulk of cells as a sample. Single-cell analyses or other techniques are needed. METHODS: In order to examine genetic stability of stem cells, we have established induced pluripotent stem cell (iPSC) lines from several patients with DNA repair-deficiency disorders, such as ataxia telangiectasia and xeroderma pigmentosum, along with healthy controls. Performing whole-exome sequencing analyses of these parental and iPSC lines, we compiled somatic mutations accumulated by the deficiency of DNA repair mechanisms. Whereas most somatic mutations cannot be detected in bulk, cell reprogramming enabled us to observe all the somatic mutations which had occurred in the cell line. Patterns of somatic mutations should be distinctive depending on which DNA repair gene is impaired. RESULTS: The comparison revealed that deficiency of ATM and XPA preferentially gives rise to indels and single-nucleotide substitutions, respectively. On the other hand, deficiency of ERCC2 caused not only single-nucleotide mutations but also many retrotranspositions of endogenous genes, which were readily identified by examining removal of introns in whole-exome sequencing. Although the number was limited, those events were also detected in healthy control samples. CONCLUSIONS: The present study exploits clonality of iPSCs to unveil somatic mutation sets that are usually hidden in bulk cell analysis. Whole-exome sequencing analysis facilitated the detection of retrotransposition mutations. The results suggest that retrotranspositions of human endogenous genes are more frequent than expected in somatic cells and that ERCC2 plays a defensive role against transposition of endogenous and exogenous DNA fragments.

Block of CDK1-dependent polyadenosine elongation of Cyclin B mRNA in metaphase-i-arrested starfish oocytes is released by intracellular pH elevation upon spawning.

Meiotic progression requires the translation of maternal mRNAs in a strict temporal order. In isolated animal oocytes, translation of maternal mRNAs containing a cytoplasmic polyadenylation element (CPE), such as cyclin B, is activated by in vitro stimulation of meiotic resumption which induces phosphorylation of CPEB (CPE-binding protein) and elongation of their polyadenosine (poly(A)) tails; whether or not this model can be applied in vivo to oocytes arrested at metaphase of meiosis I in ovaries is unknown. In this study, we found that active CDK1 (cyclin-dependent kinase 1) phosphorylated CPEB in ovarian oocytes arrested at metphase I in the starfish body cavity, but phosphorylation of CPEB was not sufficient for elongation of cyclin B poly(A) tails. Immediately after spawning, however, mRNA was polyadenylated, suggesting that an increase in intracellular pH (pHi ) upon spawning triggers the elongation of poly(A) tails. Using a cell-free system made from maturing oocytes at metaphase I, we demonstrated that polyadenylation was indeed suppressed at pH below 7.0. These results suggest that a pH-sensitive process, functioning after CPEB phosphorylation, is blocked under physiologically low pHi (<7.0) in metaphase-I-arrested oocytes. The increase in pHi (>7.0) that occurs after spawning triggers polyadenylation of cyclin B mRNA and progression into meiosis II.

Case study on the evolution of hetero-oligomer interfaces based on the differences in paralogous proteins.

We addressed the evolutionary trace of hetero-oligomer interfaces by comparing the structures of paralogous proteins; one of them is a monomer or homo-oligomer and the other is a hetero-oligomer. We found different trends in amino acid conservation pattern and hydrophobicity between homo-oligomer and hetero-oligomer. The degree of amino acid conservation in the interface of homo-oligomer has no obvious difference from that in the surface, whereas the degree of conservation is much higher in the interface of hetero-oligomer. The interface of homo-oligomer has a few very conserved residue positions, whereas the residue conservation in the interface of hetero-oligomer tends to be higher. In addition, the interface of hetero-oligomer has a tendency of being more hydrophobic compared with the one in homo-oligomer. We conjecture that these differences are related to the inherent symmetry in homo-oligomers that cannot exist in hetero-oligomers. Paucity of the structural data precludes statistical tests of these tendencies, yet the trend can be applied to the prediction of the interface of hetero-oligomer. We obtained putative interfaces of the subunits in CPSF (cleavage and polyadenylation specificity factor), one of the human pre-mRNA 3'-processing complexes. The locations of predicted interface residues were consistent with the known experimental data.