Clinical capability of modern brain tumor segmentation models.

PURPOSE: State-of-the-art automated segmentation methods achieve exceptionally high performance on the Brain Tumor Segmentation (BraTS) challenge, a dataset of uniformly processed and standardized magnetic resonance generated images (MRIs) of gliomas. However, a reasonable concern is that these models may not fare well on clinical MRIs that do not belong to the specially curated BraTS dataset. Research using the previous generation of deep learning models indicates significant performance loss on cross-institutional predictions. Here, we evaluate the cross-institutional applicability and generalzsability of state-of-the-art deep learning models on new clinical data. METHODS: We train a state-of-the-art 3D U-Net model on the conventional BraTS dataset comprising low- and high-grade gliomas. We then evaluate the performance of this model for automatic tumor segmentation of brain tumors on in-house clinical data. This dataset contains MRIs of different tumor types, resolutions, and standardization than those found in the BraTS dataset. Ground truth segmentations to validate the automated segmentation for in-house clinical data were obtained from expert radiation oncologists. RESULTS: We report average Dice scores of 0.764, 0.648, and 0.61 for the whole tumor, tumor core, and enhancing tumor, respectively, in the clinical MRIs. These means are higher than numbers reported previously on same institution and cross-institution datasets of different origin using different methods. There is no statistically significant difference when comparing the dice scores to the inter-annotation variability between two expert clinical radiation oncologists. Although performance on the clinical data is lower than on the BraTS data, these numbers indicate that models trained on the BraTS dataset have impressive segmentation performance on previously unseen images obtained at a separate clinical institution. These images differ in the imaging resolutions, standardization pipelines, and tumor types from the BraTS data. CONCLUSIONS: State-of-the-art deep learning models demonstrate promising performance on cross-institutional predictions. They considerably improve on previous models and can transfer knowledge to new types of brain tumors without additional modeling.

Hotspot exons are common targets of splicing perturbations.

High-throughput splicing assays have demonstrated that many exonic variants can disrupt splicing; however, splice-disrupting variants distribute non-uniformly across genes. We propose the existence of exons that are particularly susceptible to splice-disrupting variants, which we refer to as hotspot exons. Hotspot exons are also more susceptible to splicing perturbation through drug treatment and knock-down of RNA-binding proteins. We develop a classifier for exonic splice-disrupting variants and use it to infer hotspot exons. We estimate that 1400 exons in the human genome are hotspots. Using panels of splicing reporters, we demonstrate how the ability of an exon to tolerate a mutation is inversely proportional to the strength of its neighboring splice sites.

Tyrosine phosphorylation regulates hnRNPA2 granule protein partitioning and reduces neurodegeneration.

mRNA transport in neurons requires formation of transport granules containing many protein components, and subsequent alterations in phosphorylation status can release transcripts for translation. Further, mutations in a structurally disordered domain of the transport granule protein hnRNPA2 increase its aggregation and cause hereditary proteinopathy of neurons, myocytes, and bone. We examine in vitro hnRNPA2 granule component phase separation, partitioning specificity, assembly/disassembly, and the link to neurodegeneration. Transport granule components hnRNPF and ch-TOG interact weakly with hnRNPA2 yet partition specifically into liquid phase droplets with the low complexity domain (LC) of hnRNPA2, but not FUS LC. In vitro hnRNPA2 tyrosine phosphorylation reduces hnRNPA2 phase separation, prevents partitioning of hnRNPF and ch-TOG into hnRNPA2 LC droplets, and decreases aggregation of hnRNPA2 disease variants. The expression of chimeric hnRNPA2 D290V in Caenorhabditis elegans results in stress-induced glutamatergic neurodegeneration; this neurodegeneration is rescued by loss of tdp-1, suggesting gain-of-function toxicity. The expression of Fyn, a tyrosine kinase that phosphorylates hnRNPA2, reduces neurodegeneration associated with chimeric hnRNPA2 D290V. These data suggest a model where phosphorylation alters LC interaction specificity, aggregation, and toxicity.