Improving Low-Latency Predictions in Multi-Exit Neural Networks via Block-Dependent Losses.

As the size of a model increases, making predictions using deep neural networks (DNNs) is becoming more computationally expensive. Multi-exit neural network is one promising solution that can flexibly make anytime predictions via early exits, depending on the current test-time budget which may vary over time in practice (e.g., self-driving cars with dynamically changing speeds). However, the prediction performance at the earlier exits is generally much lower than the final exit, which becomes a critical issue in low-latency applications having a tight test-time budget. Compared to the previous works where each block is optimized to minimize the losses of all exits simultaneously, in this work, we propose a new method for training multi-exit neural networks by strategically imposing different objectives on individual blocks. The proposed idea based on grouping and overlapping strategies improves the prediction performance at the earlier exits while not degrading the performance of later ones, making our scheme to be more suitable for low-latency applications. Extensive experimental results on both image classification and semantic segmentation confirm the advantage of our approach. The proposed idea does not require any modifications in the model architecture and can be easily combined with existing strategies aiming to improve the performance of multi-exit neural networks.

Polystyrene nanoplastics affect transcriptomic and epigenomic signatures of human fibroblasts and derived induced pluripotent stem cells: Implications for human health.

Plastic pollution is increasing at an alarming rate yet the impact of this pollution on human health is poorly understood. Because human induced pluripotent stem cells (hiPSC) are frequently derived from dermal fibroblasts, these cells offer a powerful platform for the identification of molecular biomarkers of environmental pollution in human cells. Here, we describe a novel proof-of-concept for deriving hiPSC from human dermal fibroblasts deliberately exposed to polystyrene (PS) nanoplastic particles; unexposed hiPSC served as controls. In parallel, unexposed hiPSC were exposed to low and high concentrations of PS nanoparticles. Transcriptomic and epigenomic signatures of all fibroblasts and hiPSCs were defined using RNA-seq and whole genome methyl-seq, respectively. Both PS-treated fibroblasts and derived hiPSC showed alterations in expression of ESRRB and HNF1A genes and circuits involved in the pluripotency of stem cells, as well as in pathways involved in cancer, inflammatory disorders, gluconeogenesis, carbohydrate metabolism, innate immunity, and dopaminergic synapse. Similarly, the expression levels of identified key transcriptional and DNA methylation changes (DNMT3A, ESSRB, FAM133CP, HNF1A, SEPTIN7P8, and TTC34) were significantly affected in both PS-exposed fibroblasts and hiPSC. This study illustrates the power of human cellular models of environmental pollution to narrow down and prioritize the list of candidate molecular biomarkers of environmental pollution. This knowledge will facilitate the deciphering of the origins of environmental diseases.

Direct SARS-CoV-2 infection of the human inner ear may underlie COVID-19-associated audiovestibular dysfunction.

BACKGROUND: COVID-19 is a pandemic respiratory and vascular disease caused by SARS-CoV-2 virus. There is a growing number of sensory deficits associated with COVID-19 and molecular mechanisms underlying these deficits are incompletely understood. METHODS: We report a series of ten COVID-19 patients with audiovestibular symptoms such as hearing loss, vestibular dysfunction and tinnitus. To investigate the causal relationship between SARS-CoV-2 and audiovestibular dysfunction, we examine human inner ear tissue, human inner ear in vitro cellular models, and mouse inner ear tissue. RESULTS: We demonstrate that adult human inner ear tissue co-expresses the angiotensin-converting enzyme 2 (ACE2) receptor for SARS-CoV-2 virus, and the transmembrane protease serine 2 (TMPRSS2) and FURIN cofactors required for virus entry. Furthermore, hair cells and Schwann cells in explanted human vestibular tissue can be infected by SARS-CoV-2, as demonstrated by confocal microscopy. We establish three human induced pluripotent stem cell (hiPSC)-derived in vitro models of the inner ear for infection: two-dimensional otic prosensory cells (OPCs) and Schwann cell precursors (SCPs), and three-dimensional inner ear organoids. Both OPCs and SCPs express ACE2, TMPRSS2, and FURIN, with lower ACE2 and FURIN expression in SCPs. OPCs are permissive to SARS-CoV-2 infection; lower infection rates exist in isogenic SCPs. The inner ear organoids show that hair cells express ACE2 and are targets for SARS-CoV-2. CONCLUSIONS: Our results provide mechanistic explanations of audiovestibular dysfunction in COVID-19 patients and introduce hiPSC-derived systems for studying infectious human otologic disease.

Human induced pluripotent stem cells and CRISPR/Cas-mediated targeted genome editing: Platforms to tackle sensorineural hearing loss.

Hearing loss (HL) is a major global health problem of pandemic proportions. The most common type of HL is sensorineural hearing loss (SNHL) which typically occurs when cells within the inner ear are damaged. Human induced pluripotent stem cells (hiPSCs) can be generated from any individual including those who suffer from different types of HL. The development of new differentiation protocols to obtain cells of the inner ear including hair cells (HCs) and spiral ganglion neurons (SGNs) promises to expedite cell-based therapy and screening of potential pharmacologic and genetic therapies using human models. Considering age-related, acoustic, ototoxic, and genetic insults which are the most frequent causes of irreversible damage of HCs and SGNs, new methods of genome editing (GE), especially the CRISPR/Cas9 technology, could bring additional opportunities to understand the pathogenesis of human SNHL and identify novel therapies. However, important challenges associated with both hiPSCs and GE need to be overcome before scientific discoveries are correctly translated to effective and patient-safe applications. The purpose of the present review is (a) to summarize the findings from published reports utilizing hiPSCs for studies of SNHL, hence complementing recent reviews focused on animal studies, and (b) to outline promising future directions for deciphering SNHL using disruptive molecular and genomic technologies.

Postnatal expression and possible function of RANK and RANKL in the murine inner ear.

The bone encasing the inner ear, known as the otic capsule, is unique because it remodels little postnatally compared to other bones in the body. Previous studies established that osteoprotegerin (OPG) in the inner ear inhibits otic capsule remodeling. OPG acts as a decoy receptor of receptor activator of nuclear factor κB ligand (RANKL) to disrupt the interaction between RANKL and RANK, the primary regulators of bone metabolism. Here we studied the expression and function of RANK and RANKL in the murine cochlea. Using a combination of in situ hybridization, real-time quantitative RT-PCR, and western blot, we demonstrate that Rankl and Rank genes and their protein products are expressed in the intracochlear soft tissues and the otic capsule in a developmentally regulated manner. Using a culture of neonatal murine cochlear neurons, we show that the interaction between RANK and RANKL inhibits neurite outgrowth in these neurons, and is associated with upregulation of NOGO-A expression. Taken together, our results suggest that, in addition to regulating otic capsule bone remodeling, RANK and RANKL expressed by intracochlear soft tissues may also regulate spiral ganglion neuron function by affecting neurite outgrowth.

Flexible transparent film heaters using a ternary composite of silver nanowire, conducting polymer, and conductive oxide.

Scientific and technological advances in transparent conductive electrodes improve the heating performance of flexible transparent film heaters (TFHs), which can be utilized for various applications as defrosters and heaters. To achieve high performance as well as practical TFHs, several conditions, such as high optical transmittance, low electrical resistance, heating uniformity, and operational stability in various environmental conditions should be satisfied. However, due to the trade-offs between optical transmittance and electrical resistance, it is not easy to fulfill all the requirements concurrently. Here we report flexible TFHs using a ternary composite of silver nanowire (AgNW), conducting polymer (i.e., poly[3,4-ethylenedioxythiophene]:polystyrene sulfonate [PEDOT:PSS]), and a thin conductive oxide (i.e., indium tin oxide [ITO]) layer, exhibiting higher performance in terms of the maximum heating temperature (>110 °C), operational stability, mechanical flexibility, and optical transmittance (95% at 550 nm), compared to pristine AgNW-based TFHs. We also demonstrated the stable operation of the AgNW-PEDOT:PSS/ITO TFHs soaked in water, showing excellent environmental stability. To analyse the fundamental mechanisms for the improved performance of the AgNW-PEDOT:PSS/ITO TFHs, we investigated the progress of joule heating using a device simulator, and found that the improvement originated not only from reduced electrical resistance but also from enhanced heat dissipation with PEDOT:PSS and ITO. We anticipate that our analysis and results will be helpful for further development of practical flexible TFHs.

Monoallelic gene targeting in hypoblast stem cells reveals X chromosome inactivation.

We recently isolated hypoblast stem cells (HypoSC), which are related to embryonic stem (ES) cells. ES cells efficiently perform homologous recombination (HR) and lack X chromosome inactivation (Xi), but it is unknown whether the same applies to HypoSC. Using the X-linked hypoxanthine phosphoribosyl transferase (HPRT) gene, we find that HypoSC perform HR with similar frequency as ES cells. Monoallelic targeting in female HypoSC eliminated HPRT gene expression, implying epigenetic inactivation of the other allele. Although density-induced differentiation complicated selection, the targeted clones maintained their original properties. These results will facilitate targeted genetic manipulation of HypoSC and the study of Xi.