Comprehensive Study on Molecular Supervised Learning with Graph Neural Networks.

This work considers strategies to develop accurate and reliable graph neural networks (GNNs) for molecular property predictions. Prediction performance of GNNs is highly sensitive to the change in various parameters due to the inherent challenges in molecular machine learning, such as a deficient amount of data samples and bias in data distribution. Comparative studies with well-designed experiments are thus important to clearly understand which GNNs are powerful for molecular supervised learning. Our work presents a number of ablation studies along with a guideline to train and utilize GNNs for both molecular regression and classification tasks. First, we validate that using both atomic and bond meta-information improves the prediction performance in the regression task. Second, we find that the graph isomorphism hypothesis proposed by [Xu, K.; et al How powerful are graph neural networks? 2018, arXiv:1810.00826. arXiv.org e-Print archive. https://arxiv.org/abs/1810.00826] is valid for the regression task. Surprisingly, however, the findings above do not hold for the classification tasks. Beyond the study on model architectures, we test various regularization methods and Bayesian learning algorithms to find the best strategy to achieve a reliable classification system. We demonstrate that regularization methods penalizing predictive entropy might not give well-calibrated probability estimation, even though they work well in other domains, and Bayesian learning methods are capable of developing reliable prediction systems. Furthermore, we argue the importance of Bayesian learning in virtual screening by showing that well-calibrated probability estimation may lead to a higher success rate.

Development of a common platform for the noninvasive prenatal diagnosis of X-linked diseases.

OBJECTIVE: The aim of this study was to develop a common targeted massively parallel sequencing platform for the noninvasive prenatal diagnosis (NIPD) of multiple X-linked diseases. METHOD: The custom capture probe was designed to target 33 genes and recombination hotspots. We tested the carrier mother and male proband pair of 6 families. Plasma DNA of the pregnant carrier mother was collected at different gestational weeks and sequenced. The fetal genotype of each family was determined by estimating the imbalance between the 2 maternal haplotypes constructed using a common custom-designed platform. RESULTS: The targeted sequencing of the maternal, proband, and fetal genomic DNAs and maternal plasma DNAs resulted in uniform coverage across the target region. Three to 5 recombination points were observed in each sample. However, these recombination points did not affect the haplotype dosage analysis for fetal genotype prediction. Consequently, all fetal genotypes in the 6 families obtained from haplotype dosage analysis of maternal plasma sequencing data were predicted correctly. CONCLUSIONS: Since a single platform that covers multiple diseases may prevent the need for disease-specific probes for the NIPD of individual disorders, this approach may provide a practical advantage for clinically implementing the NIPD of X-linked diseases.

Application of sperm selection using hyaluronic acid binding in intracytoplasmic sperm injection cycles: a sibling oocyte study.

The purpose of this study was to investigate whether sperm selection by hyaluronic acid (HA) binding could improve fertilization rate and embryo quality in intracytoplasmic sperm injection (ICSI) cycles. Two hundred nineteen oocytes obtained from eighteen women were injected with either HA-bound (n = 107) or conventionally selected spermatozoa (n = 112) in a randomized way. All of the participants were infertile couples who had normal sperm parameters but low fertilization rate in previous in vitro fertilization (IVF) cycle (n = 5) or experienced multiple IVF failures (n = 13). Lower fertilization (75.7% vs 83.0%) and cleavage rate on day 2 (72.9% vs 83.0%) was observed in oocytes injected with HA-bound spermatozoa than the conventional group, but the difference was not significant. Significantly lower cleavage rate was observed on day 3 in HA group (56.0% vs 69.6%, P = 0.038). Blastocyst formation rate and the number of transferred embryos were similar in both groups. In multiple IVF failure patients, significantly reduced fertilization rate (71.8% vs 85.3%, P = 0.046) and cleavage rate on day 2 (70.4% vs 85.3%, P = 0.029) and day 3 (53.5% vs 77.3%, P = 0.002) were noticed in HA group. Five women achieved pregnancy continuing more than 12 weeks after transfer (27.8%). Success of ICSI was not related with the number of embryos fertilized by HA-bound spermatozoa. Application of ICSI by sperm selection using HA binding is not helpful in couples with repeated poor fertilization or implantation despite normal sperm parameters.

CONFIGURE: A pipeline for identifying context specific regulatory modules from gene expression data and its application to breast cancer.

BACKGROUND: Gene expression data is widely used for identifying subtypes of diseases such as cancer. Differentially expressed gene analysis and gene set enrichment analysis are widely used for identifying biological mechanisms at the gene level and gene set level, respectively. However, the results of differentially expressed gene analysis are difficult to interpret and gene set enrichment analysis does not consider the interactions among genes in a gene set. RESULTS: We present CONFIGURE, a pipeline that identifies context specific regulatory modules from gene expression data. First, CONFIGURE takes gene expression data and context label information as inputs and constructs regulatory modules. Then, CONFIGURE makes a regulatory module enrichment score (RMES) matrix of enrichment scores of the regulatory modules on samples using the single-sample GSEA method. CONFIGURE calculates the importance scores of the regulatory modules on each context to rank the regulatory modules. We evaluated CONFIGURE on the Cancer Genome Atlas (TCGA) breast cancer RNA-seq dataset to determine whether it can produce biologically meaningful regulatory modules for breast cancer subtypes. We first evaluated whether RMESs are useful for differentiating breast cancer subtypes using a multi-class classifier and one-vs-rest binary SVM classifiers. The multi-class and one-vs-rest binary classifiers were trained using the RMESs as features and outperformed baseline classifiers. Furthermore, we conducted literature surveys on the basal-like type specific regulatory modules obtained by CONFIGURE and showed that highly ranked modules were associated with the phenotypes of basal-like type breast cancers. CONCLUSIONS: We showed that enrichment scores of regulatory modules are useful for differentiating breast cancer subtypes and validated the basal-like type specific regulatory modules by literature surveys. In doing so, we found regulatory module candidates that have not been reported in previous literature. This demonstrates that CONFIGURE can be used to predict novel regulatory markers which can be validated by downstream wet lab experiments. We validated CONFIGURE on the breast cancer RNA-seq dataset in this work but CONFIGURE can be applied to any gene expression dataset containing context information.

Clinicopathological characteristics of miscarriages featuring placental massive perivillous fibrin deposition.

INTRODUCTION: Massive perivillous fibrin deposition (MPFD) is frequently associated with detrimental pregnancy outcomes, and extensive perivillous fibrin deposition results in severe placental dysfunction and loss of maternofetal interface. Unfortunately, the fundamental pathogenesis of MPFD remains unknown, and systematic analyses of MPFD in miscarriage is lacking. We analyzed the frequency and clinicopathological characteristics of MPFD in first trimester miscarriages. METHODS: We analyzed a consecutive series of miscarriages (n = 582) gathered between March 2012 and June 2016. MPFD was classified as fibrin-type (f-MPFD) and matrix-type (m-MPFD) by immunostaining for fibrin and collagen type IV. The control group consisted of miscarriage cases (MC, n = 18) that were matched to f-MPFD with normal chromosome (f-MPFD-nc) for number of previous miscarriages and placental chromosomal status. RESULTS: MPFD was identified in 2.7% of miscarriages. f-MPFD was associated with recurrent abortions. Compared with miscarriages without fibrin deposition, MPFD cases had higher proportion of those with normal placental chromosome (69.2% vs. 27.4%, P < 0.005) and higher frequency of villous syncytiotrophoblast C4d deposition (73.3% vs. 33.9%, P < 0.005). All C4d(+) f-MPFD patients had more than three recurrent miscarriages, whereas C4d(-) f-MPFD patients had no history of recurrent miscarriage (P < 0.05). Patients with f-MPFD-nc had significantly higher HLA PRA immunopositivity rate than did MC patients (P = 0.005). DISCUSSION: MPFD was more common in miscarriages than in preterm and term pregnancies. Placental massive fibrin-type fibrinoid deposition and villous C4d immunoreactivity were associated with recurrent miscarriage.

Clinical experience with multiplex ligation-dependent probe amplification for microdeletion syndromes in prenatal diagnosis: 7522 pregnant Korean women.

BACKGROUND: Conventional cytogenetic analysis using G-band karyotyping has been the method of choice for prenatal diagnosis, accurately detecting chromosomal abnormalities larger than 5 Mb. However, the method is inefficient for detecting the submicroscopic deletions and duplications that are associated with malformations and mental retardation. This study evaluated the results of the multiplex ligation-dependent probe amplification (MLPA) P245 assay used for prenatal diagnosis in cases with unusual ultrasonographic findings or specifically where parents wanted to be tested. The objective was to compare the results from MLPA with those from conventional cytogenetic testing in order to determine their concordance and the additional diagnostic yield of MLPA over G-band karyotyping. RESULTS: Of the 7522 prenatal cases analyzed, 124 were found to have genomic imbalances (1.6%). Of those 124 cases, 41 had gene loss (33.6%), and 83 had gene gain (66.4%). Most of the cases with genomic imbalances (64.5%) showed no abnormal karyotype. In particular, all cases with a 4p16.3 deletion (Wolf-Hirschhorn syndrome) showed an abnormal karyotype, whereas all of those with a 22q11-13 deletion showed a normal karyotype. In most of the cases with pathogenic deletions, the indication for invasive prenatal testing was an increase in the nuchal translucency (NT) alone (51.2%). Other indications observed in the remaining cases were abnormal serum screening markers (14.6%), other ultrasonographic findings (9.8%), pregnancy through in vitro fertilization and fertility assistance (9.8%), and advanced maternal age(2.4%). CONCLUSIONS: These results show that for fetuses with an enlarged NT or abnormal ultrasonographic findings and normal conventional karyotype, additional genetic investigation like molecular testing would be for identifying the microscopic genomic aberrations (microdeletions, microduplications) responsible for syndromic associations including structural anomalies and mental retardation.

Expression of the genes for peroxisome proliferator-activated receptor-γ, cyclooxygenase-2, and proinflammatory cytokines in granulosa cells from women with polycystic ovary syndrome.

OBJECTIVE: To identify differences in the expression of the genes for peroxisome proliferator-activated receptor (PPAR)-γ, cyclooxygenase (COX)-2, and the proinflammatory cytokines interleukin (IL)-6 and tumor necrosis factor (TNF)-α in granulosa cells (GCs) from polycystic ovary syndrome (PCOS) patients and controls undergoing controlled ovarian stimulation. METHODS: Nine patients with PCOS and six controls were enrolled in this study. On the day of oocyte retrieval, GCs were collected from pooled follicular fluid. Total mRNA was extracted from GCs. Reverse transcription was performed and gene expression levels were quantified by realtime quantitative polymerase chain reaction. RESULTS: There were no significant differences in age, body mass index, and total gonadotropin dose, except for the ratio of luteinizing hormone to follicle-stimulating hormone between the PCOS and control groups. PPAR-γ and COX-2 mRNA was significantly downregulated in the GCs of PCOS women compared with controls (p=0.034 and p=0.018, respectively), but the expression of IL-6 and TNF-α mRNA did not show significant differences. No significant correlation was detected between the expression of these mRNA sequences and clinical characteristics, including the number of retrieved oocytes, oocyte maturity, cleavage, or the good embryo rate. Positive correlations were found among the PPAR-γ, COX-2, IL-6, and TNF-α mRNA levels. CONCLUSION: Our data may provide novel clues regarding ovarian GC dysfunction in PCOS, and indirectly provide evidence that the effect of PPAR-γ agonists in PCOS might result from alterations in the ovarian follicular environment. Further studies with a larger sample size are required to confirm these proposals.

Fetal gender determination and BclI polymorphism using nucleated erythrocytes in maternal blood.

This study demonstrated determination of fetal gender from nucleated red blood cells (NRBCs) in maternal blood and attempted to apply prenatal diagnosis of hemophilia A using BclI DNA polymorphism. Venous blood was drawn from 20 pregnant women, and NRBCs were recovered by magnetic activated cell sorting and anti-GPA (glycophorin A) immunostaining. After microdissector isolation of the NRBCs, primer extension preamplification (PEP) and nested PCR of the amelogenin gene were performed to determine fetal gender. We also performed PEP and nested PCR of BclI polymorphism to verify the validity of prenatal diagnosis of hemophilia A. DNA amplification was achieved in 107 cells (51.9%) and fetal gender determined with 65.0% accuracy. Unfortunately, we could not verify the validity within the scope of this study. However, in a larger number of cases that are informative in BclI polymorphism, we will be able to identify patients affected by hemophilia A using fetal NRBCs in maternal blood.

Diffuse decidual leucocytoclastic necrosis and NK cell aggregates: two distinct features of miscarriage.

AIMS: Primary histopathology of miscarriage remains undetermined in the majority of cases. This study was conducted to determine histological characteristics pertinent to miscarriage. METHODS: The study groups were composed of elective abortions (n=29) and miscarriages (n=45) comprised of chromosomally normal (n=15) and abnormal cases (n=30). Immunohistochemistry was done against CD3, CD8, TIA-1 and CD56. RESULTS: Two histological features--diffuse decidual leucocytoclastic necrosis (DDLN) and decidual natural killer cell aggregates (NKCA)--were relatively common in miscarriages. The frequencies of DDLN and NKCA were different between the groups (p<0.05 and p<0.05, respectively). DDLN was found in 13.8% (4/29) of elective abortions, while it was observed in 60.0% (9/15) and 23.3% (7/30) of chromosomally normal and abnormal miscarriages, respectively. DDLN was more frequent in chromosomally normal miscarriages than in elective abortions (p=0.004). NKCA was present in 13.8% (4/29) of elective abortions, while being found in 33.3% (5/15) and 43.3% (13/30) of chromosomally normal and abnormal miscarriages, respectively. NKCA was more frequent in chromosomally abnormal miscarriages than in elective abortions (p=0.020). CONCLUSIONS: The findings strongly suggest that defective placentation and abnormal maternal immune response are associated with miscarriage. DDLN and NKCA seem to have diagnostic values in the pathological evaluation of miscarriage.

Placental C4d deposition is a feature of defective placentation: observations in cases of preeclampsia and miscarriage.

Placental C4d deposition is frequent in preeclampsia, and shallow placentation is a characteristic of both preeclampsia and miscarriage. This study was conducted to determine the relationship among placental C4d, maternal human leukocyte antigen (HLA) antibodies, and placental pathology in preeclampsia and miscarriage cases. The patient population (N = 104) included those with (1) preterm preeclampsia with fetal growth restriction (PE-FGR; n = 21), (2) preterm preeclampsia (PE; n = 20), (3) spontaneous preterm delivery (sPTD; n = 39), and (4) miscarriage (n = 24). C4d immunohistochemistry was performed, and the presence of maternal plasma HLA antibodies was examined. C4d staining of the syncytiotrophoblast was more frequent in PE-FGR patients (76.2 %) than in PE (10.0 %; p < 0.001) and sPTD (2.6 %; p < 0.001) patients. Maternal HLA antibody-positive rate was not different among the study groups. There was a significant correlation between C4d immunoreactivity and placental pathology consistent with maternal vascular underperfusion (p < 0.001) but not with maternal HLA antibody status. In miscarriages, the positive rates of C4d, HLA class I, and HLA class II antibodies were 58.3, 25.0, and 12.5 %, respectively. There was no correlation between the presence of maternal HLA class I or II antibodies and placental C4d immunoreactivity. This study confirms frequent placental C4d deposition in preeclampsia with fetal growth restriction and miscarriage. The association between placental C4d deposition and pathological findings of maternal vascular underperfusion suggests that C4d staining of the syncytiotrophoblast is a consequence of defective placentation rather than of a specific maternal immune response against fetal HLA. The study also demonstrates the usefulness of C4d as a biomarker of placentas at risk.

Placental C4d as a common feature of chromosomally normal and abnormal miscarriages.

Placental C4d deposition is a feature of classical complement pathway activation and has been documented in various obstetrical settings. However, it is unknown whether placental C4d deposition is present in miscarriages and its frequency is different between chromosomally normal and abnormal miscarriages. This study was conducted to assess villous C4d deposition in miscarriages and to determine whether its frequency is different between chromosomally normal and abnormal miscarriages. Tissue samples (N = 58) of elective abortions (n = 20), miscarriages with normal chromosomes (n = 15), trisomy 16 (n = 13), and trisomy 22 (n = 10) were analyzed. Immunohistochemical staining for C4d and CD138 was done. Placental C4d deposition was defined as linear C4d immunoreactivity along the syncytiotrophoblast. Placental C4d immunoreactivity was detected in 73.3 % (11/15) and 56.5 % (13/23) of miscarriages with normal chromosomes and trisomy cases, respectively, while it was found in 5 % (1/20) of elective abortions (p < 0.05). Placental C4d deposition was more frequent in recurrent miscarriages (previous spontaneous abortion ≥2) than in sporadic miscarriages (76.5 vs. 30.0 %; p = 0.001). Chronic deciduitis was observed in 20.0 % (3/15) and 30.4 % (7/23) of miscarriages with normal chromosomes and trisomy cases, respectively, but not in elective abortions (p = 0.07 and 0.01, for each). The frequencies of C4d deposition (46.2 vs. 70.0 %) and chronic deciduitis (38.5 vs. 20.0 %) were not also different between trisomy 16 and trisomy 22 cases. Placental C4d deposition is a prominent feature of miscarriages regardless of their chromosomal status. The overall findings suggest that complement-mediated placental injury is a common pathology of miscarriage with diagnostic values in routine pathology practice.

The relationship between the number of CGG repeats and serum level of anti-Müllerian hormone in women without FMR1 premutation.

OBJECTIVE: To investigate the relationship between the number of CGG repeats in the fragile X mental retardation 1 (FMR1) gene and serum levels of anti-Müllerian hormone (AMH) in Korean infertility patients without premutation. STUDY DESIGN: A retrospective study of 228 infertile women who received fertility treatment in a single private in vitro fertilization (IVF) clinic from May 2010 to August 2012 was performed. Serum FSH and AMH were measured on menstrual day 3 and the number of CGG repeats was evaluated. RESULTS: The mean age of the study population was 33.3±3.8 years. No significant correlation was observed between CGG repeat count in both alleles and the serum FSH, AMH or multiples of median (MoM) of AMH in whole study subjects. In women with age ≥35 years, however, there was an increasing tendency in the MoM of AMH with increasing number of CGG repeats in allele 2 (R(2)=0.075, p=0.008). This correlation was not observed in patients aged less than 35 years. CONCLUSION: We observed a positive correlation between MoM of AMH and number of CGG repeats in allele 2 in women aged over 35 years. Our findings are discordant with other reports, and therefore further studies are needed to determine whether this discrepancy is due to ethnic differences.

Outcomes of preimplantation genetic diagnosis using either zona drilling with acidified Tyrode's solution or partial zona dissection.

OBJECTIVE: To review the outcomes of preimplantation genetic diagnosis (PGD) using zona drilling with acid Tyrode's solution (chemical zona pellucida drilling, chemical ZD) and those of partial zona dissection (PZD). METHODS: Clinical outcomes of seventy-one couples undergoing 85 PGD cycles from January 2005 to December 2010 were included. Blastocyst formation and the hatching rate, clinical pregnancy rate, ongoing pregnancy rate, implantation rate, and fetal gender ratio of the PZD and chemical ZD groups were compared. RESULTS: Application of PZD resulted in a significantly higher rate of clinical pregnancy (40.7% vs. 15.4%, p=0.022), ongoing pregnancy (35.6% vs. 11.5%, p=0.023), and implantation (18.1% vs. 5.7%, p=0.007) compared with chemical ZD. Among non-transferred embryos, the rate of blastocyst formation on day 5 (49.1% vs. 39.5%, p=0.016) and hatching on day 6 (47.2% vs. 26.5%, p<0.001) were also significantly higher in the PZD group. CONCLUSION: The mechanical zona dissection method showed better outcomes than chemical ZD in terms of the blastocyst development and pregnancy rate. In this study, the fact that chemical ZD was conducted in different period from mechanical method should be considered in interpreting the result.

Identification of origin of unknown derivative chromosomes by array-based comparative genomic hybridization using pre- and postnatal clinical samples.

Microarray-based comparative genomic hybridization (array CGH) is a high-resolution and comprehensive method for detecting both genome-wide and chromosome-specific copy-number imbalance. We have developed an array CGH analysis system (consisting of an array CGH chip plus its exclusive analysis software) for constitutional genetic diagnosis and have evaluated the suitability of our system for molecular diagnosis using pre- and postnatal clinical samples. In a blind study, each of the 264 sample karyotypes identified by array CGH analysis was consistent with that identified by traditional karyotype analysis--with one exception, case (47, XXX)--and we were able to identify origins, such as small supernumerary marker chromosomes, which cannot be determined by conventional cytogenetics. We also acquired very accurate, fast and reliable results using a diminutive amount of clinical samples. Taken together, the array CGH platform developed in this study is a rapid, powerful and sensitive technology for pre- and postnatal diagnosis using a very small amount of clinical sample.