Generating post-hoc explanations for Skip-gram-based node embeddings by identifying important nodes with bridgeness.

Node representation learning in a network is an important machine learning technique for encoding relational information in a continuous vector space while preserving the inherent properties and structures of the network. Recently, unsupervised node embedding methods such as DeepWalk (Perozzi et al., 2014), LINE (Tang et al., 2015), struc2vec (Ribeiro et al., 2017), PTE (Tang et al., 2015), UserItem2vec (Wu et al., 2020), and RWJBG (Li et al., 2021) have emerged from the Skip-gram model (Mikolov et al., 2013) and perform better performance in several downstream tasks such as node classification and link prediction than the existing relational models. However, providing post-hoc explanations of unsupervised embeddings remains a challenging problem because of the lack of explanation methods and theoretical studies applicable for embeddings. In this paper, we first show that global explanations to the Skip-gram-based embeddings can be found by computing bridgeness under a spectral cluster-aware local perturbation. Moreover, a novel gradient-based explanation method, which we call GRAPH-wGD, is proposed that allows the top-q global explanations about learned graph embedding vectors more efficiently. Experiments show that the ranking of nodes by scores using GRAPH-wGD is highly correlated with true bridgeness scores. We also observe that the top-q node-level explanations selected by GRAPH-wGD have higher importance scores and produce more changes in class label prediction when perturbed, compared with the nodes selected by recent alternatives, using five real-world graphs.

Prenatal diagnosis of zellweger syndrome by fetal MRI: a case report.

Zellweger Syndrome (ZS) is a rare peroxisomal disorder also referred to as cerebrohepatorenal syndrome. ZS is an autosomal recessive disease often manifesting in the neonatal period with profound dysfunction of the central nervous system, liver and kidneys. Prenatal diagnosis of this syndrome is infrequent with imaging findings on fetal MRI rarely illustrated in the literature. This case highlights the pivotal role fetal MRI can play in identifying subtle features of the disease that are difficult to visualize on prenatal ultrasound. It is important for pediatric radiologists to be familiar with the most common imaging features of ZS on fetal MRI to expedite the diagnosis and help facilitate appropriate prenatal counseling.

Variational Random Function Model for Network Modeling.

Link prediction is a fundamental problem in network modeling. A family of link prediction approaches is to treat network data as an exchangeable array whose entries can be explained by random functions (e.g., block models and Gaussian processes) over latent node factors. Despite their powerful ability in modeling missing links, these models tend to have a large computational complexity and thus are hard to deal with large networks. To address this problem, we develop a novel variational random function model by defining latent Gaussian processes on exchangeable arrays. This model not only inherits the ability of Gaussian process to describe the nonlinear interactions between nodes, but also enjoys significant reduction on computational complexity. To further make the model scalable to large network data, we develop an efficient key-value-free strategy under the map-reduce framework to tremendously reduce the inference time. Experimental results on large network data have demonstrated both the efficacy and efficiency of the proposed method over state-of-the-arts methods in network modeling.

Usefulness of cellular text messaging for improving adherence among adolescents and young adults with systemic lupus erythematosus.

OBJECTIVE: In a cohort of 70 patients with childhood-onset systemic lupus erythematosus (cSLE): to determine the baseline adherence to medications and visits; to investigate the effects of cellular text messaging reminders (CTMR) on adherence to clinic visits; and to study the influence of CTMR on adherence to use of hydroxychloroquine (HCQ). METHODS: CTMR were sent to 70 patients prior to clinic visits for 14 months. A subgroup of patients were evaluated for medication adherence to HCQ: 19 patients receiving CTMR prior to each scheduled HCQ dose were compared to 22 patients randomized to standard of care education about HCQ. Visit adherence was measured using administrative databases. Pharmacy refill information, self-report of adherence, and HCQ blood levels were utilized to monitor medication adherence to HCQ. Sufficient adherence to visits or HCQ was defined as estimates > 80%. Disease activity was primarily monitored with the Systemic Lupus Erythematosus Disease Activity Index. RESULTS: At baseline, 32% of patients were sufficiently adherent to HCQ, and 81% to clinic visits. Visit adherence improved significantly by > 80% among those who were nonadherent to clinic visits at the baseline CTMR (p = 0.01). CTMR did not influence adherence to HCQ over time. CONCLUSION: Patients with cSLE were only modestly adherent to HCQ and clinic visits. CTMR may be effective for improving visit adherence among adolescents and young adults with cSLE, but it does not improve adherence to HCQ.

Fascicular anatomy of human femoral nerve: implications for neural prostheses using nerve cuff electrodes.

Clinical interventions to restore standing or stepping by using nerve cuff stimulation require a detailed knowledge of femoral nerve neuroanatomy. We harvested eight femoral nerves with all distal branches and characterized the branching patterns and diameters. The fascicular representation of each distal nerve was identified and traced proximally to create fascicle maps of the compound femoral nerve in four cadaver specimens. Distal nerves were consistently represented as individual fascicles or distinct groups of fascicles in the compound femoral nerve. Branch-free length of the compound femoral nerve was 1.50 +/- 0.47 cm (mean +/- standard deviation). Compound femoral nerve cross sections were noncircular with major and minor diameters of 10.50 +/- 1.52 mm and 2.30 +/- 0.63 mm, respectively. In vivo intraoperative measurements in six subjects were consistent with cadaver results. Selective stimulation of individual muscles innervated by the femoral nerve may therefore be possible with a single neural prosthesis able to selectively stimulate individual groups of fascicles.

Temporal aspects of spatial task performance during intermittent hypoxia in the rat: evidence for neurogenesis.

Intermittent hypoxia (IH) during sleep, such as occurs in obstructive sleep apnea, leads to degenerative changes in the hippocampus, and is associated with spatial learning deficits in the adult rat. We report that in Sprague-Dawley rats the initial IH-induced impairments in spatial learning are followed by a partial functional recovery over time, despite continuing IH exposure. These functional changes coincide with initial decreases in basal neurogenesis as shown by the number of positively colabelled cells for BrdU and neurofilament in the dentate gyrus of the hippocampus, and are followed by increased expression of neuronal progenitors and mature neurons (nestin and BrdU-neurofilament positively labelled cells, respectively). In contrast, no changes occurred during the course of IH exposures in the expression of the synaptic proteins synaptophysin, SNAP25, and drebrin. Collectively, these findings indicate that the occurrence of IH during the lights on period results in a biphasic pattern of neurogenesis in the hippocampus of adult rats, and may account for the observed partial recovery of spatial function.

Respiratory effects of gestational intermittent hypoxia in the developing rat.

Intermittent hypoxia (IH), one of the hallmarks of obstructive sleep apnea, occurs more frequently during pregnancy. We hypothesized that IH may lead to persistent postnatal changes in respiratory responses to acute hypoxia and may also lead to adverse effects on spatial function learning as revealed by the Morris water maze. To examine this issue, time-pregnant Sprague-Dawley rats were exposed to IH and room air (IHRA; 21 and 10% O2 alternations every 90 seconds) or to normoxia (RARA) until delivery. Ventilatory and metabolic responses to a 20-minute acute hypoxic challenge (10% O2) were conducted at postnatal ages 5, 10, 15, and 30 days. In addition, spatial task learning was assessed in the water maze at 1 and 4 months of age. Normoxic ventilation was higher at all time points in IHRA rats than in RARA rats (p < 0.01). Peak hypoxic ventilatory responses were attenuated in IHRA rats at 5 days of age and hypoxic ventilatory depression was accentuated at this age as well. However, ventilatory equivalents (minute ventilation/oxygen consumption) revealed significant reductions in peak hypoxic ventilatory responses of IHRA rats and hypoxic ventilatory depression at all postnatal ages (p < 0.01). Acquisition and retention of a spatial task were similar in the IHRA and RARA groups at both 1 and 4 months of age. We conclude that gestational intermittent hypoxia elicits long-lasting alterations in the control of breathing. We postulate that such IH-induced respiratory plasticity may create selective vulnerability to hypoxia during development.

Impaired spatial learning and hyperactivity in developing rats exposed to intermittent hypoxia.

Obstructive sleep apnea (OSA) is a frequent medical condition and is associated with cognitive impairments in adults and with hyperactivity and decreased school performance in children. In an adult rodent model, intermittent hypoxia (IH), such as occurs in OSA, is associated with neurodegenerative changes in the hippocampus and cortex and with spatial learning deficits. Because a unique developmental window of neural vulnerability to IH is present, we hypothesized that exposure to IH throughout the vulnerable ages would result in increased behavioral impairments in the juvenile rat. Rat pups were therefore exposed to either room air or IH beginning at postnatal (PN) d 10 until PN d 30. Learning and memory were assessed via a standard place-training version of the Morris water maze beginning at PN d 25. Locomotor activity was assessed on PN d 29 and 30. Pups exposed to IH displayed significant spatial learning impairments, and exposed male rats but not female rats displayed increased locomotor activity in the open field. Collectively, these findings indicate that exposure to IH at an age that corresponds to the peak incidence of OSA in children induces substantial learning impairment and gender-dependent behavioral hyperactivity in the juvenile rat. We postulate that this novel experimental model may allow for future exploration of mechanisms underlying the neurobehavioral deficits of children with OSA.