More Like this

1. OrbitN: A Symplectic Integrator for Planetary Systems Dominated by a Central Mass—Insight into Long-term Solar System Chaos

   https://doi.org/10.3847/1538-3881/acd63b  

   Zeebe, Richard E. ( June 2023 , The Astronomical Journal)

   Reliable studies of the long-term dynamics of planetary systems require numerical integrators that are accurate and fast. The challenge is often formidable because the chaotic nature of many systems requires relative numerical error bounds at or close to machine precision (∼10⁻¹⁶, double-precision arithmetic); otherwise, numerical chaos may dominate over physical chaos. Currently, the speed/accuracy demands are usually only met by symplectic integrators. For example, the most up-to-date long-term astronomical solutions for the solar system in the past (widely used in, e.g., astrochronology and high-precision geological dating) have been obtained using symplectic integrators. However, the source codes of these integrators are unavailable. Here I present the symplectic integratororbitN(lean version 1.0) with the primary goal of generating accurate and reproducible long-term orbital solutions for near-Keplerian planetary systems (here the solar system) with a dominant massM₀. Among other features,orbitN-1.0includesM₀’s quadrupole moment, a lunar contribution, and post-Newtonian corrections (1PN) due toM₀(fast symplectic implementation). To reduce numerical round-off errors, Kahan compensated summation was implemented. I useorbitNto provide insight into the effect of various processes on the long-term chaos in the solar system. Notably, 1PN corrections have the opposite effect on chaoticity/stability on a 100 Myr versus Gyr timescale. For the current application,orbitNis about as fast as or faster (factor 1.15–2.6) than comparable integrators, depending on hardware.¹¹

   The orbitN source code (C) is available athttp://github.com/rezeebe/orbitN.

    

   more » « less
2. Variation in the location and timing of experimental severing demonstrates that the persistent rhizome serves multiple functions in a clonal forest understorey herb

   https://doi.org/10.1111/1365-2435.13970  

   Eisen, Katherine E. ; Siegmund, Gregor‐Fausto ; Watson, Maxine A. ; Geber, Monica A. ( December 2021 , Functional Ecology)

   In clonal plants, persistent rhizomes can serve multiple purposes, including resource storage, modulation of heterogeneous resource distributions, maintenance of bud banks and promotion of recovery from disturbance. Clonal plants are commonly long‐lived and, in temperate zones, often exhibit organ preformation. Thus, investigations of how the timing of disturbance to the rhizome affects plant performance must occur over multiple growing seasons, but these types of studies are rare.

   We conducted a field experiment to examine how the persistent rhizome supports the existing shoot, new ramet production and recovery from damage using mayapplePodophyllum peltatum(Berberidaceae), a common herbaceous perennial of low‐light forest understories in Eastern North America. Mayapple maintains a long‐lived rhizome and exhibits a developmentally programmed seasonal pattern of resource transport and new ramet initiation. We varied both the position and timing of rhizome severing in rhizome systems with terminal sexual or vegetative shoots, and tracked plants for 2 years following severing.

   The location and timing of severing affected both plant persistence (production of new shoots) and performance (leaf area), with effects differing for new shoots that developed at the front versus the back of the rhizome system. Across years, severing location and past years’ shoot size influenced plant persistence and performance, while the effect of timing of severing diminished. Initial sexual status had little effect on rhizome system response that was not accounted for by initial leaf area. Severing generally led to the establishment of two independent rhizome systems. Relative to unmanipulated control systems, these two systems had more total leaf area, but less average leaf area per system.

   Synthesis. Our results point to the rhizome as a resource integrator that affects plant responses to disturbance immediately following damage and in subsequent growing seasons. Rhizome bud age and/or subtending rhizome size, and developmental programme influence responses to disturbance. While the effects of experimental disturbance on plant performance decreased 2 years after disturbance, further long‐term investigation is needed to fully understand the demographic consequences of damage to persistent rhizomes.

   A freePlain Language Summarycan be found within the Supporting Information of this article.

    

   more » « less
3. The Temple University Digital Pathology Corpus: The Breast Tissue Subset

   https://doi.org/10.1109/SPMB52430.2021.9672275  

   Wevodau, Z. ; Doshna, B. ; Jhala, N. ; Akhtar, I. ; Obeid, I. ; Picone, J. ( December 2021 , Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB))

   Obeid, I. (Ed.)

   The Neural Engineering Data Consortium (NEDC) is developing the Temple University Digital Pathology Corpus (TUDP), an open source database of high-resolution images from scanned pathology samples [1], as part of its National Science Foundation-funded Major Research Instrumentation grant titled “MRI: High Performance Digital Pathology Using Big Data and Machine Learning” [2]. The long-term goal of this project is to release one million images. We have currently scanned over 100,000 images and are in the process of annotating breast tissue data for our first official corpus release, v1.0.0. This release contains 3,505 annotated images of breast tissue including 74 patients with cancerous diagnoses (out of a total of 296 patients). In this poster, we will present an analysis of this corpus and discuss the challenges we have faced in efficiently producing high quality annotations of breast tissue. It is well known that state of the art algorithms in machine learning require vast amounts of data. Fields such as speech recognition [3], image recognition [4] and text processing [5] are able to deliver impressive performance with complex deep learning models because they have developed large corpora to support training of extremely high-dimensional models (e.g., billions of parameters). Other fields that do not have access to such data resources must rely on techniques in which existing models can be adapted to new datasets [6]. A preliminary version of this breast corpus release was tested in a pilot study using a baseline machine learning system, ResNet18 [7], that leverages several open-source Python tools. The pilot corpus was divided into three sets: train, development, and evaluation. Portions of these slides were manually annotated [1] using the nine labels in Table 1 [8] to identify five to ten examples of pathological features on each slide. Not every pathological feature is annotated, meaning excluded areas can include focuses particular to these labels that are not used for training. A summary of the number of patches within each label is given in Table 2. To maintain a balanced training set, 1,000 patches of each label were used to train the machine learning model. Throughout all sets, only annotated patches were involved in model development. The performance of this model in identifying all the patches in the evaluation set can be seen in the confusion matrix of classification accuracy in Table 3. The highest performing labels were background, 97% correct identification, and artifact, 76% correct identification. A correlation exists between labels with more than 6,000 development patches and accurate performance on the evaluation set. Additionally, these results indicated a need to further refine the annotation of invasive ductal carcinoma (“indc”), inflammation (“infl”), nonneoplastic features (“nneo”), normal (“norm”) and suspicious (“susp”). This pilot experiment motivated changes to the corpus that will be discussed in detail in this poster presentation. To increase the accuracy of the machine learning model, we modified how we addressed underperforming labels. One common source of error arose with how non-background labels were converted into patches. Large areas of background within other labels were isolated within a patch resulting in connective tissue misrepresenting a non-background label. In response, the annotation overlay margins were revised to exclude benign connective tissue in non-background labels. Corresponding patient reports and supporting immunohistochemical stains further guided annotation reviews. The microscopic diagnoses given by the primary pathologist in these reports detail the pathological findings within each tissue site, but not within each specific slide. The microscopic diagnoses informed revisions specifically targeting annotated regions classified as cancerous, ensuring that the labels “indc” and “dcis” were used only in situations where a micropathologist diagnosed it as such. Further differentiation of cancerous and precancerous labels, as well as the location of their focus on a slide, could be accomplished with supplemental immunohistochemically (IHC) stained slides. When distinguishing whether a focus is a nonneoplastic feature versus a cancerous growth, pathologists employ antigen targeting stains to the tissue in question to confirm the diagnosis. For example, a nonneoplastic feature of usual ductal hyperplasia will display diffuse staining for cytokeratin 5 (CK5) and no diffuse staining for estrogen receptor (ER), while a cancerous growth of ductal carcinoma in situ will have negative or focally positive staining for CK5 and diffuse staining for ER [9]. Many tissue samples contain cancerous and non-cancerous features with morphological overlaps that cause variability between annotators. The informative fields IHC slides provide could play an integral role in machine model pathology diagnostics. Following the revisions made on all the annotations, a second experiment was run using ResNet18. Compared to the pilot study, an increase of model prediction accuracy was seen for the labels indc, infl, nneo, norm, and null. This increase is correlated with an increase in annotated area and annotation accuracy. Model performance in identifying the suspicious label decreased by 25% due to the decrease of 57% in the total annotated area described by this label. A summary of the model performance is given in Table 4, which shows the new prediction accuracy and the absolute change in error rate compared to Table 3. The breast tissue subset we are developing includes 3,505 annotated breast pathology slides from 296 patients. The average size of a scanned SVS file is 363 MB. The annotations are stored in an XML format. A CSV version of the annotation file is also available which provides a flat, or simple, annotation that is easy for machine learning researchers to access and interface to their systems. Each patient is identified by an anonymized medical reference number. Within each patient’s directory, one or more sessions are identified, also anonymized to the first of the month in which the sample was taken. These sessions are broken into groupings of tissue taken on that date (in this case, breast tissue). A deidentified patient report stored as a flat text file is also available. Within these slides there are a total of 16,971 total annotated regions with an average of 4.84 annotations per slide. Among those annotations, 8,035 are non-cancerous (normal, background, null, and artifact,) 6,222 are carcinogenic signs (inflammation, nonneoplastic and suspicious,) and 2,714 are cancerous labels (ductal carcinoma in situ and invasive ductal carcinoma in situ.) The individual patients are split up into three sets: train, development, and evaluation. Of the 74 cancerous patients, 20 were allotted for both the development and evaluation sets, while the remain 34 were allotted for train. The remaining 222 patients were split up to preserve the overall distribution of labels within the corpus. This was done in hope of creating control sets for comparable studies. Overall, the development and evaluation sets each have 80 patients, while the training set has 136 patients. In a related component of this project, slides from the Fox Chase Cancer Center (FCCC) Biosample Repository (https://www.foxchase.org/research/facilities/genetic-research-facilities/biosample-repository -facility) are being digitized in addition to slides provided by Temple University Hospital. This data includes 18 different types of tissue including approximately 38.5% urinary tissue and 16.5% gynecological tissue. These slides and the metadata provided with them are already anonymized and include diagnoses in a spreadsheet with sample and patient ID. We plan to release over 13,000 unannotated slides from the FCCC Corpus simultaneously with v1.0.0 of TUDP. Details of this release will also be discussed in this poster. Few digitally annotated databases of pathology samples like TUDP exist due to the extensive data collection and processing required. The breast corpus subset should be released by November 2021. By December 2021 we should also release the unannotated FCCC data. We are currently annotating urinary tract data as well. We expect to release about 5,600 processed TUH slides in this subset. We have an additional 53,000 unprocessed TUH slides digitized. Corpora of this size will stimulate the development of a new generation of deep learning technology. In clinical settings where resources are limited, an assistive diagnoses model could support pathologists’ workload and even help prioritize suspected cancerous cases. ACKNOWLEDGMENTS This material is supported by the National Science Foundation under grants nos. CNS-1726188 and 1925494. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. REFERENCES [1] N. Shawki et al., “The Temple University Digital Pathology Corpus,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York City, New York, USA: Springer, 2020, pp. 67 104. https://www.springer.com/gp/book/9783030368432. [2] J. Picone, T. Farkas, I. Obeid, and Y. Persidsky, “MRI: High Performance Digital Pathology Using Big Data and Machine Learning.” Major Research Instrumentation (MRI), Division of Computer and Network Systems, Award No. 1726188, January 1, 2018 – December 31, 2021. https://www. isip.piconepress.com/projects/nsf_dpath/. [3] A. Gulati et al., “Conformer: Convolution-augmented Transformer for Speech Recognition,” in Proceedings of the Annual Conference of the International Speech Communication Association (INTERSPEECH), 2020, pp. 5036-5040. https://doi.org/10.21437/interspeech.2020-3015. [4] C.-J. Wu et al., “Machine Learning at Facebook: Understanding Inference at the Edge,” in Proceedings of the IEEE International Symposium on High Performance Computer Architecture (HPCA), 2019, pp. 331–344. https://ieeexplore.ieee.org/document/8675201. [5] I. Caswell and B. Liang, “Recent Advances in Google Translate,” Google AI Blog: The latest from Google Research, 2020. [Online]. Available: https://ai.googleblog.com/2020/06/recent-advances-in-google-translate.html. [Accessed: 01-Aug-2021]. [6] V. Khalkhali, N. Shawki, V. Shah, M. Golmohammadi, I. Obeid, and J. Picone, “Low Latency Real-Time Seizure Detection Using Transfer Deep Learning,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB), 2021, pp. 1 7. https://www.isip. piconepress.com/publications/conference_proceedings/2021/ieee_spmb/eeg_transfer_learning/. [7] J. Picone, T. Farkas, I. Obeid, and Y. Persidsky, “MRI: High Performance Digital Pathology Using Big Data and Machine Learning,” Philadelphia, Pennsylvania, USA, 2020. https://www.isip.piconepress.com/publications/reports/2020/nsf/mri_dpath/. [8] I. Hunt, S. Husain, J. Simons, I. Obeid, and J. Picone, “Recent Advances in the Temple University Digital Pathology Corpus,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB), 2019, pp. 1–4. https://ieeexplore.ieee.org/document/9037859. [9] A. P. Martinez, C. Cohen, K. Z. Hanley, and X. (Bill) Li, “Estrogen Receptor and Cytokeratin 5 Are Reliable Markers to Separate Usual Ductal Hyperplasia From Atypical Ductal Hyperplasia and Low-Grade Ductal Carcinoma In Situ,” Arch. Pathol. Lab. Med., vol. 140, no. 7, pp. 686–689, Apr. 2016. https://doi.org/10.5858/arpa.2015-0238-OA. 

   more » « less
4. Unveiling the warm and dense ISM in z  > 6 quasar host galaxies via water vapor emission

   https://doi.org/10.1051/0004-6361/202243406  

   Pensabene, A. ; van der Werf, P. ; Decarli, R. ; Bañados, E. ; Meyer, R. A. ; Riechers, D. ; Venemans, B. ; Walter, F. ; Weiß, A. ; Brusa, M. ; et al ( November 2022 , Astronomy & Astrophysics)

   Water vapor (H₂O) is one of the brightest molecular emitters after carbon monoxide (CO) in galaxies with high infrared (IR) luminosity, allowing us to investigate the warm and dense phase of the interstellar medium (ISM) where star formation occurs. However, due to the complexity of its radiative spectrum, H₂O is not frequently exploited as an ISM tracer in distant galaxies. Therefore, H₂O studies of the warm and dense gas at high-zremain largely unexplored. In this work, we present observations conducted with the Northern Extended Millimeter Array (NOEMA) toward threez > 6 IR-bright quasarsJ2310+1855,J1148+5251, andJ0439+1634targeted in their multiple para- and ortho-H₂O transitions (3₁₂ − 3₀₃, 1₁₁ − 0₀₀, 2₂₀ − 2₁₁, and 4₂₂ − 4₁₃), as well as their far-IR (FIR) dust continuum. By combining our data with previous measurements from the literature, we estimated the dust masses and temperatures, continuum optical depths, IR luminosities, and star formation rates (SFR) from the FIR continuum. We modeled the H₂O lines using the MOLPOP-CEP radiative transfer code, finding that water vapor lines in our quasar host galaxies are primarily excited in the warm, dense (with a gas kinetic temperature and density ofT_kin = 50 K,n_H2 ∼ 10^4.5 − 10⁵ cm⁻³) molecular medium with a water vapor column density ofN_H2O ∼ 2 × 10¹⁷ − 3 × 10¹⁸ cm⁻³. High-JH₂O lines are mainly radiatively pumped by the intense optically-thin far-IR radiation field associated with a warm dust component at temperatures ofT_dust ∼ 80 − 190 K that account for < 5 − 10% of the total dust mass. In the case of J2310+1855, our analysis points to a relatively high value of the continuum optical depth at 100 μm (τ₁₀₀ ∼ 1). Our results are in agreement with expectations based on the H₂O spectral line energy distribution of local and high-zultra-luminous IR galaxies and active galactic nuclei (AGN). The analysis of the Boltzmann diagrams highlights the interplay between collisions and IR pumping in populating the high H₂O energy levels and it allows us to directly compare the excitation conditions in the targeted quasar host galaxies. In addition, the observations enable us to sample the high-luminosity part of the H₂O–total-IR (TIR) luminosity relations (L_H2O − L_TIR). Overall, our results point to supralinear trends that suggest H₂O–TIR relations are likely driven by IR pumping, rather than the mere co-spatiality between the FIR continuum- and line-emitting regions. The observedL_H2O/L_TIRratios in ourz > 6 quasars do not show any strong deviations with respect to those measured in star-forming galaxies and AGN at lower redshifts. This supports the notion that H₂O can be likely used to trace the star formation activity buried deep within the dense molecular clouds.

    

   more » « less
5. Genome analyses of the new model protist Euplotes vannus focusing on genome rearrangement and resistance to environmental stressors

   https://doi.org/10.1111/1755-0998.13023  

   Chen, Xiao ; Jiang, Yaohan ; Gao, Feng ; Zheng, Weibo ; Krock, Timothy J. ; Stover, Naomi A. ; Lu, Chao ; Katz, Laura A. ; Song, Weibo ( June 2019 , Molecular Ecology Resources)

   As a model organism for studies of cell and environmental biology, the free‐living and cosmopolitan ciliateEuplotes vannusshows intriguing features like dual genome architecture (i.e., separate germline and somatic nuclei in each cell/organism), “gene‐sized” chromosomes, stop codon reassignment, programmed ribosomal frameshifting (PRF) and strong resistance to environmental stressors. However, the molecular mechanisms that account for these remarkable traits remain largely unknown. Here we report a combined analysis of de novo assembled high‐quality macronuclear (MAC; i.e., somatic) and partial micronuclear (MIC; i.e., germline) genome sequences forE. vannus, and transcriptome profiling data under varying conditions. The results demonstrate that: (a) the MAC genome contains more than 25,000 complete “gene‐sized” nanochromosomes (~85 Mb haploid genome size) with the N50 ~2.7 kb; (b) although there is a high frequency of frameshifting at stop codons UAA and UAG, we did not observe impaired transcript abundance as a result of PRF in this species as has been reported for other euplotids; (c) the sequence motif 5′‐TA‐3′ is conserved at nearly all internally‐eliminated sequence (IES) boundaries in the MIC genome, and chromosome breakage sites (CBSs) are duplicated and retained in the MAC genome; (d) by profiling the weighted correlation network of genes in the MAC under different environmental stressors, including nutrient scarcity, extreme temperature, salinity and the presence of ammonia, we identified gene clusters that respond to these external physical or chemical stimulations, and (e) we observed a dramatic increase in HSP70 gene transcription under salinity and chemical stresses but surprisingly, not under temperature changes; we link this temperature‐resistance to the evolved loss of temperature stress‐sensitive elements in regulatory regions. Together with the genome resources generated in this study, which are available online atEuplotes vannusGenome Database (http://evan.ciliate.org), these data provide molecular evidence for understanding the unique biology of highly adaptable microorganisms.

    

   more » « less