Deep Learning for Visual Localization and Mapping: A Survey.

Deep-learning-based localization and mapping approaches have recently emerged as a new research direction and receive significant attention from both industry and academia. Instead of creating hand-designed algorithms based on physical models or geometric theories, deep learning solutions provide an alternative to solve the problem in a data-driven way. Benefiting from the ever-increasing volumes of data and computational power on devices, these learning methods are fast evolving into a new area that shows potential to track self-motion and estimate environmental models accurately and robustly for mobile agents. In this work, we provide a comprehensive survey and propose a taxonomy for the localization and mapping methods using deep learning. This survey aims to discuss two basic questions: whether deep learning is promising for localization and mapping, and how deep learning should be applied to solve this problem. To this end, a series of localization and mapping topics are investigated, from the learning-based visual odometry and global relocalization to mapping, and simultaneous localization and mapping (SLAM). It is our hope that this survey organically weaves together the recent works in this vein from robotics, computer vision, and machine learning communities and serves as a guideline for future researchers to apply deep learning to tackle the problem of visual localization and mapping.

SenseFi: A library and benchmark on deep-learning-empowered WiFi human sensing.

Over the recent years, WiFi sensing has been rapidly developed for privacy-preserving, ubiquitous human-sensing applications, enabled by signal processing and deep-learning methods. However, a comprehensive public benchmark for deep learning in WiFi sensing, similar to that available for visual recognition, does not yet exist. In this article, we review recent progress in topics ranging from WiFi hardware platforms to sensing algorithms and propose a new library with a comprehensive benchmark, SenseFi. On this basis, we evaluate various deep-learning models in terms of distinct sensing tasks, WiFi platforms, recognition accuracy, model size, computational complexity, and feature transferability. Extensive experiments are performed whose results provide valuable insights into model design, learning strategy, and training techniques for real-world applications. In summary, SenseFi is a comprehensive benchmark with an open-source library for deep learning in WiFi sensing research that offers researchers a convenient tool to validate learning-based WiFi-sensing methods on multiple datasets and platforms.

Learning Selective Sensor Fusion for State Estimation.

Autonomous vehicles and mobile robotic systems are typically equipped with multiple sensors to provide redundancy. By integrating the observations from different sensors, these mobile agents are able to perceive the environment and estimate system states, e.g., locations and orientations. Although deep learning (DL) approaches for multimodal odometry estimation and localization have gained traction, they rarely focus on the issue of robust sensor fusion--a necessary consideration to deal with noisy or incomplete sensor observations in the real world. Moreover, current deep odometry models suffer from a lack of interpretability. To this extent, we propose SelectFusion, an end-to-end selective sensor fusion module that can be applied to useful pairs of sensor modalities, such as monocular images and inertial measurements, depth images, and light detection and ranging (LIDAR) point clouds. Our model is a uniform framework that is not restricted to specific modality or task. During prediction, the network is able to assess the reliability of the latent features from different sensor modalities and to estimate trajectory at both scale and global pose. In particular, we propose two fusion modules--a deterministic soft fusion and a stochastic hard fusion--and offer a comprehensive study of the new strategies compared with trivial direct fusion. We extensively evaluate all fusion strategies both on public datasets and on progressively degraded datasets that present synthetic occlusions, noisy and missing data, and time misalignment between sensors, and we investigate the effectiveness of the different fusion strategies in attending the most reliable features, which in itself provides insights into the operation of the various models.

An effective AKT inhibitor-PARP inhibitor combination therapy for recurrent ovarian cancer.

BACKGROUND: Although the use of PARP inhibitor has received considerable amount of attention in ovarian cancer, PARP inhibitor resistance still emerges with disease progression. PI3K/AKT pathway inhibitors have been proposed to synergize with PARP inhibition to slow tumor growth, but the exact molecular mechanisms are still elusive. METHODS: Utilizing tumor samples from recurrent EOC patients with platinum resistance and prior PARP inhibitor use, Mini PDX and PDX models were established to study the anti-tumor effect of AKT inhibitor (LAE003) and LAE003/PARP inhibitor (Olaparib) in combination. Five ovarian cancer cell lines were treated with Olaparib or LAE003 or in combination in vitro. Cell viability and apoptosis rate were measured after the treatments. Combination index by the Chou-Talalay was used to evaluate in vitro combination effect of Olaparib and LAE003. The protein expression level of PARP1 and PAR was measured by Western blot in cell lines and by immunohistochemistry in PDX tumor tissues. RESULTS: Tumor cells from two out of five platinum-resistant ovarian cancer patients previously treated with PARP inhibitor were sensitive to AKT inhibition in Mini-PDX study. Inhibition of AKT further increased the response of tumor cells to Olaparib in a PDX model derived from a recurrent platinum-resistant ovarian cancer patient. Additive anti-proliferation effect of LAE003 and Olaparib was also observed in three ovarian cancer cell lines with high PARP1 protein level. Interestingly, mechanism study revealed that AKT inhibition decreased PARP enzyme activity as measured by PAR level and/or reduced PARP1 protein level in the tumor cell lines and PDX tumor tissues, which may explain the observed combined anti-tumor effect of LAE003 and Olaparib. CONCLUSION: Collectively, our results suggest that the combination of AKT inhibitor and PARP inhibitor could be a viable approach for clinical testing in recurrent ovarian cancer patients.

DynaNet: Neural Kalman Dynamical Model for Motion Estimation and Prediction.

Dynamical models estimate and predict the temporal evolution of physical systems. State-space models (SSMs) in particular represent the system dynamics with many desirable properties, such as being able to model uncertainty in both the model and measurements, and optimal (in the Bayesian sense) recursive formulations, e.g., the Kalman filter. However, they require significant domain knowledge to derive the parametric form and considerable hand tuning to correctly set all the parameters. Data-driven techniques, e.g., recurrent neural networks, have emerged as compelling alternatives to SSMs with wide success across a number of challenging tasks, in part due to their impressive capability to extract relevant features from rich inputs. They, however, lack interpretability and robustness to unseen conditions. Thus, data-driven models are hard to be applied in safety-critical applications, such as self-driving vehicles. In this work, we present DynaNet, a hybrid deep learning and time-varying SSM, which can be trained end-to-end. Our neural Kalman dynamical model allows us to exploit the relative merits of both SSM and deep neural networks. We demonstrate its effectiveness in the estimation and prediction on a number of physically challenging tasks, including visual odometry, sensor fusion for visual-inertial navigation, and motion prediction. In addition, we show how DynaNet can indicate failures through investigation of properties, such as the rate of innovation (Kalman gain).

Circulating Cell-Free DNA-Based Detection of Tumor Suppressor Gene Copy Number Loss and Its Clinical Implication in Metastatic Prostate Cancer.

Current liquid biopsy assays lack sufficient sensitivity to detect copy number loss, which limits the interrogation of critical tumor suppressor gene deletions during cancer progression and treatment. Here we describe a liquid biopsy assay with improved sensitivity for detection of copy number loss in blood samples with low levels of circulating tumor DNA, and demonstrate its utility by profiling PTEN, RB1, and TP53 genetic loss in metastatic prostate cancer patients.

The Unified Medical Language System SPECIALIST Lexicon and Lexical Tools: Development and applications.

Natural language processing (NLP) plays a vital role in modern medical informatics. It converts narrative text or unstructured data into knowledge by analyzing and extracting concepts. A comprehensive lexical system is the foundation to the success of NLP applications and an essential component at the beginning of the NLP pipeline. The SPECIALIST Lexicon and Lexical Tools, distributed by the National Library of Medicine as one of the Unified Medical Language System Knowledge Sources, provides an underlying resource for many NLP applications. This article reports recent developments of 3 key components in the Lexicon. The core NLP operation of Unified Medical Language System concept mapping is used to illustrate the importance of these developments. Our objective is to provide generic, broad coverage and a robust lexical system for NLP applications. A novel multiword approach and other planned developments are proposed.

Spell checker for consumer language (CSpell).

Objective: Automated understanding of consumer health inquiries might be hindered by misspellings. To detect and correct various types of spelling errors in consumer health questions, we developed a distributable spell-checking tool, CSpell, that handles nonword errors, real-word errors, word boundary infractions, punctuation errors, and combinations of the above. Methods: We developed a novel approach of using dual embedding within Word2vec for context-dependent corrections. This technique was used in combination with dictionary-based corrections in a 2-stage ranking system. We also developed various splitters and handlers to correct word boundary infractions. All correction approaches are integrated to handle errors in consumer health questions. Results: Our approach achieves an F1 score of 80.93% and 69.17% for spelling error detection and correction, respectively. Discussion: The dual-embedding model shows a significant improvement (9.13%) in F1 score compared with the general practice of using cosine similarity with word vectors in Word2vec for context ranking. Our 2-stage ranking system shows a 4.94% improvement in F1 score compared with the best 1-stage ranking system. Conclusion: CSpell improves over the state of the art and provides near real-time automatic misspelling detection and correction in consumer health questions. The software and the CSpell test set are available at https://umlslex.nlm.nih.gov/cSpell.

Upregulation of CD11b and CD86 through LSD1 inhibition promotes myeloid differentiation and suppresses cell proliferation in human monocytic leukemia cells.

LSD1 (Lysine Specific Demethylase1)/KDM1A (Lysine Demethylase 1A), a flavin adenine dinucleotide (FAD)-dependent histone H3K4/K9 demethylase, sustains oncogenic potential of leukemia stem cells in primary human leukemia cells. However, the pro-differentiation and anti-proliferation effects of LSD1 inhibition in acute myeloid leukemia (AML) are not yet fully understood. Here, we report that small hairpin RNA (shRNA) mediated LSD1 inhibition causes a remarkable transcriptional activation of myeloid lineage marker genes (CD11b/ITGAM and CD86), reduction of cell proliferation and decrease of clonogenic ability of human AML cells. Cell surface expression of CD11b and CD86 is significantly and dynamically increased in human AML cells upon sustained LSD1 inhibition. Chromatin immunoprecipitation and quantitative PCR (ChIP-qPCR) analyses of histone marks revealed that there is a specific increase of H3K4me2 modification and an accompanied increase of H3K4me3 modification at the respective CD11b and CD86 promoter region, whereas the global H3K4me2 level remains constant. Consistently, inhibition of LSD1 in vivo significantly blocks tumor growth and induces a prominent increase of CD11b and CD86. Taken together, our results demonstrate the anti-tumor properties of LSD1 inhibition on human AML cell line and mouse xenograft model. Our findings provide mechanistic insights into the LSD1 functions in controlling both differentiation and proliferation in AML.

Discovery and Molecular Basis of a Diverse Set of Polycomb Repressive Complex 2 Inhibitors Recognition by EED.

Polycomb repressive complex 2 (PRC2), a histone H3 lysine 27 methyltransferase, plays a key role in gene regulation and is a known epigenetics drug target for cancer therapy. The WD40 domain-containing protein EED is the regulatory subunit of PRC2. It binds to the tri-methylated lysine 27 of the histone H3 (H3K27me3), and through which stimulates the activity of PRC2 allosterically. Recently, we disclosed a novel PRC2 inhibitor EED226 which binds to the K27me3-pocket on EED and showed strong antitumor activity in xenograft mice model. Here, we further report the identification and validation of four other EED binders along with EED162, the parental compound of EED226. The crystal structures for all these five compounds in complex with EED revealed a common deep pocket induced by the binding of this diverse set of compounds. This pocket was created after significant conformational rearrangement of the aromatic cage residues (Y365, Y148 and F97) in the H3K27me3 binding pocket of EED, the width of which was delineated by the side chains of these rearranged residues. In addition, all five compounds interact with the Arg367 at the bottom of the pocket. Each compound also displays unique features in its interaction with EED, suggesting the dynamics of the H3K27me3 pocket in accommodating the binding of different compounds. Our results provide structural insights for rational design of novel EED binder for the inhibition of PRC2 complex activity.

Split luciferase-based biosensors for characterizing EED binders.

The EED (embryonic ectoderm development) subunit of the Polycomb repressive complex 2 (PRC2) plays an important role in the feed forward regulation of the PRC2 enzymatic activity. We recently identified a new class of allosteric PRC2 inhibitors that bind to the H3K27me3 pocket of EED. Multiple assays were developed and used to identify and characterize this type of PRC2 inhibitors. One of them is a genetically encoded EED biosensor based on the EED[G255D] mutant and the split firefly luciferase. This EED biosensor can detect the compound binding in the transfected cells and in the in vitro biochemical assays. Compared to other commonly used cellular assays, the EED biosensor assay has the advantage of shorter compound incubation with cells. The in vitro EED biosensor is much more sensitive than other label-free biophysical assays (e.g. DSF, ITC). Based on the crystal structure, the DSF data as well as the biosensor assay data, it's most likely that compound-induced increase in the luciferase activity of the EED[G255D] biosensor results from the decreased non-productive interactions between the EED subdomain and other subdomains within the biosensor construct. This new insight of the mechanism might help to broaden the use of the split luciferase based biosensors.

Enhanced LexSynonym Acquisition for Effective UMLS Concept Mapping.

Concept mapping is important in natural language processing (NLP) for bioinformatics. The UMLS Metathesaurus provides a rich synonym thesaurus and is a popular resource for concept mapping. Query expansion using synonyms for subterm substitutions is an effective technique to increase recall for UMLS concept mapping. Synonyms used to substitute subterms are called element synonyms. The completeness and quality of both element synonyms and the UMLS synonym thesaurus is the key to success in such applications. The Lexical Systems Group (LSG) has developed a new system for element synonym acquisition based on new enhanced requirements and design for better performance. The results show: 1) A 36.71 times growth of synonyms in the Lexicon (lexSynonym) in the 2017 release; 2) Improvements of concept mapping for recall and F1 with similar precision using the lexSynonym.2017 as element synonyms due to the broader coverage and better quality.

Histone methyltransferase SETDB1 regulates liver cancer cell growth through methylation of p53.

SETDB1 is a histone H3K9 methyltransferase that has a critical role in early development. It is located within a melanoma susceptibility locus and facilitates melanoma formation. However, the mechanism by which SETDB1 regulates tumorigenesis remains unknown. Here we report the molecular interplay between SETDB1 and the well-known hotspot gain-of-function (GOF) TP53 R249S mutation. We show that in hepatocellular carcinoma (HCC) SETDB1 is overexpressed with moderate copy number gain, and GOF TP53 mutations including R249S associate with this overexpression. Inactivation of SETDB1 in HCC cell lines bearing the R249S mutation suppresses cell growth. The TP53 mutation status renders cancer cells dependent on SETDB1. Moreover, SETDB1 forms a complex with p53 and catalyses p53K370 di-methylation. SETDB1 attenuation reduces the p53K370me2 level, which subsequently leads to increased recognition and degradation of p53 by MDM2. Together, we provide both genetic and biochemical evidence for a mechanism by which SETDB1 regulates cancer cell growth via methylation of p53.

SETDB1 modulates PRC2 activity at developmental genes independently of H3K9 trimethylation in mouse ES cells.

SETDB1, a histone methyltransferase responsible for methylation of histone H3 lysine 9 (H3K9), is involved in maintenance of embryonic stem (ES) cells and early embryonic development of the mouse. However, how SETDB1 regulates gene expression during development is largely unknown. Here, we characterized genome-wide SETDB1 binding and H3K9 trimethylation (H3K9me3) profiles in mouse ES cells and uncovered two distinct classes of SETDB1 binding sites, termed solo and ensemble peaks. The solo peaks were devoid of H3K9me3 and enriched near developmental regulators while the ensemble peaks were associated with H3K9me3. A subset of the SETDB1 solo peaks, particularly those near neural development-related genes, was found to be associated with Polycomb Repressive Complex 2 (PRC2) as well as PRC2-interacting proteins JARID2 and MTF2. Genetic deletion of Setdb1 reduced EZH2 binding as well as histone 3 lysine 27 (H3K27) trimethylation level at SETDB1 solo peaks and facilitated neural differentiation. Furthermore, we found that H3K27me3 inhibits SETDB1 methyltransferase activity. The currently identified reciprocal action between SETDB1 and PRC2 reveals a novel mechanism underlying ES cell pluripotency and differentiation regulation.

Divergent behaviors and underlying mechanisms of cell migration and invasion in non-metastatic T24 and its metastatic derivative T24T bladder cancer cell lines.

Previous studies on cancer cell invasion were primarily focused on its migration because these two events were often considered biologically equivalent. Here we found that T24T cells exhibited higher invasion but lower migration abilities than T24 cells. Expression of Rho-GDPases was much lower and expression of SOD2 was much higher in T24T cells than those in T24 cells. Indeed, knockdown of SOD2 in T24T cells can reverse the cell migration but without affecting cell invasion. We also found that SOD2 inhibited the JNK/c-Jun cascade, and the inhibition of c-Jun activation by ectopic expression of TAM67 impaired Rho-GDPases expression and cell migration in T24T shSOD2 cells. Further, we found that Sp1 can upregulate SOD2 transcription in T24T cells. Importantly, matrix metalloproteinase-2 (MMP-2) was overexpressed in T24T and participated in increasing its invasion, and MMP-2 overexpression was mediated by increasing nuclear transport of nucleolin, which enhanced mmp-2 mRNA stability. Taken together, our study unravels an inverse relationship between cell migration and invasion in human bladder cancer T24T cells and suggests a novel mechanism underlying the divergent roles of SOD2 and MMP-2 in regulating metastatic behaviors of human bladder T24T in cell migration and invasion.

Absolute quantification of histone PTM marks by MRM-based LC-MS/MS.

The N-terminal tails of core histones harbor the sites of numerous post-translational modifications (PTMs) with important roles in the regulation of chromatin structure and function. Profiling histone PTM marks provides data that help understand the epigenetics events in cells and their connections with cancer and other diseases. Our previous study demonstrated that specific derivatization of histone peptides by NHS propionate significantly improved their chromatographic performance on reversed phase columns for LC/MS analysis. As a step forward, we recently developed a multiple reaction monitoring (MRM) based LC-MS/MS method to analyze 42 targeted histone peptides. By using stable isotopic labeled peptides as internal standards that are spiked into the reconstituted solutions, this method allows to measure absolute concentration of the tryptic peptides of H3 histone proteins extracted from cancer cell lines. The method was thoroughly validated for the accuracy and reproducibility through analyzing recombinant histone proteins and cellular samples. The linear dynamic range of the MRM assays was achieved in 3 orders of magnitude from 1 nM to 1 µM for all targeted peptides. Excellent intrabatch and interbatch reproducibility (<15% CV) was obtained. This method has been used to study translocated NSD2 (a histone lysine methyltransferase that catalyzes the histone lysine 36 methylation) function with its overexpression in KMS11 multiple myeloma cells. From the results we have successfully quantitated both individual and combinatorial histone marks in parental and NSD2 selective knockout KMS11 cells.

Nuclear receptor NR4A1 promotes breast cancer invasion and metastasis by activating TGF-ß signalling.

In advanced cancers, the TGF-ß pathway acts as an oncogenic factor and is considered to be a therapeutic target. Here using a genome-wide cDNA screen, we identify nuclear receptor NR4A1 as a strong activator of TGF-ß signalling. NR4A1 promotes TGF-ß/SMAD signalling by facilitating AXIN2-RNF12/ARKADIA-induced SMAD7 degradation. NR4A1 interacts with SMAD7 and AXIN2, and potently and directly induces AXIN2 expression. Whereas loss of NR4A1 inhibits TGF-ß-induced epithelial-to-mesenchymal transition and metastasis, slight NR4A1 ectopic expression stimulates metastasis in a TGF-ß-dependent manner. Importantly, inflammatory cytokines potently induce NR4A1 expression, and potentiate TGF-ß-mediated breast cancer cell migration, invasion and metastasis in vitro and in vivo. Notably, NR4A1 expression is elevated in breast cancer patients with high immune infiltration and its expression weakly correlates with phosphorylated SMAD2 levels, and is an indicator of poor prognosis. Our results uncover inflammation-induced NR4A1 as an important determinant for hyperactivation of pro-oncogenic TGF-ß signalling in breast cancer.

NSD2 is recruited through its PHD domain to oncogenic gene loci to drive multiple myeloma.

Histone lysine methyltransferase NSD2 (WHSC1/MMSET) is overexpressed frequently in multiple myeloma due to the t(4;14) translocation associated with 15% to 20% of cases of this disease. NSD2 has been found to be involved in myelomagenesis, suggesting it may offer a novel therapeutic target. Here we show that NSD2 methyltransferase activity is crucial for clonogenicity, adherence, and proliferation of multiple myeloma cells on bone marrow stroma in vitro and that NSD2 is required for tumorigenesis of t(4;14)+ but not t(4;14)- multiple myeloma cells in vivo. The PHD domains in NSD2 were important for its cellular activity and biological function through recruiting NSD2 to its oncogenic target genes and driving their transcriptional activation. By strengthening its disease linkage and deepening insights into its mechanism of action, this study provides a strategy to therapeutically target NSD2 in multiple myeloma patients with a t(4;14) translocation.

TRAF4 promotes TGF-ß receptor signaling and drives breast cancer metastasis.

TGF-ß signaling is a therapeutic target in advanced cancers. We identified tumor necrosis factor receptor-associated factor 4 (TRAF4) as a key component mediating pro-oncogenic TGF-ß-induced SMAD and non-SMAD signaling. Upon TGF-ß stimulation, TRAF4 is recruited to the active TGF-ß receptor complex, where it antagonizes E3 ligase SMURF2 and facilitates the recruitment of deubiquitinase USP15 to the TGF-ß type I receptor (TßRI). Both processes contribute to TßRI stabilization on the plasma membrane and thereby enhance TGF-ß signaling. In addition, the TGF-ß receptor-TRAF4 interaction triggers Lys 63-linked TRAF4 polyubiquitylation and subsequent activation of the TGF-ß-activated kinase (TAK)1. TRAF4 is required for efficient TGF-ß-induced migration, epithelial-to-mesenchymal transition, and breast cancer metastasis. Elevated TRAF4 expression correlated with increased levels of phosphorylated SMAD2 and phosphorylated TAK1 as well as poor prognosis among breast cancer patients. Our results demonstrate that TRAF4 can regulate the TGF-ß pathway and is a key determinant in breast cancer pathogenesis.