Search

Background:This study used machine learning to develop a 3-year lung cancer risk prediction model with large real-world data in a mostly younger population.Methods:Over 4.7 million individuals, aged 45 to 65 years with no history of any cancer or lung cancer screening, diagnostic, or treatment procedures, with an outpatient visit in 2013 were identified in Optum's de-identified Electronic Health Record (EHR) dataset. A least absolute shrinkage and selection operator model was fit using all available data in the 365 days prior. Temporal validation was assessed with recent data. External validation was assessed with data from Mercy Health Systems EHR and Optum's de-identified Clinformatics Data Mart Database. Racial inequities in model discrimination were assessed with xAUCs.Results:The model AUC was 0.76. Top predictors included age, smoking, race, ethnicity, and diagnosis of chronic obstructive pulmonary disease. The model identified a high-risk group with lung cancer incidence 9 times the average cohort incidence, representing 10% of patients with lung cancer. Model performed well temporally and externally, while performance was reduced for Asians and Hispanics.Conclusions:A high-dimensional model trained using big data identified a subset of patients with high lung cancer risk. The model demonstrated transportability to EHR and claims data, while underscoring the need to assess racial disparities when using machine learning methods.Impact:This internally and externally validated real-world data-based lung cancer prediction model is available on an open-source platform for broad sharing and application. Model integration into an EHR system could minimize physician burden by automating identification of high-risk patients.

BACKGROUND: Adverse pregnancy outcomes (APO) contribute to higher risk of maternal cerebrovascular disease, but longitudinal data that include APO and stroke timing are lacking. We hypothesized that APO are associated with younger age at first stroke, with a stronger relationship in those with >1 pregnancy with APO. METHODS: We analyzed longitudinal Finnish nationwide health registry data from the FinnGen Study. We included women who gave birth after 1969 when the hospital discharge registry was established. We defined APO as a pregnancy affected by gestational hypertension, preeclampsia, eclampsia, preterm birth, small for gestational age infant, or placental abruption. We defined stroke as first hospital admission for ischemic stroke or nontraumatic intracerebral or subarachnoid hemorrhage, excluding stroke during pregnancy or within 1 year postpartum. We used Kaplan-Meier survival curves and multivariable-adjusted Cox and generalized linear models to assess the relationship between APO and future stroke. RESULTS: We included 144 306 women with a total of 316 789 births in the analysis sample, of whom 17.9% had at least 1 pregnancy with an APO and 2.9% experienced an APO in ≥2 pregnancies. Women with APO had more comorbidities including obesity, hypertension, heart disease, and migraine. Median age at first stroke was 58.3 years in those with no APO, 54.8 years in those with 1 APO, and 51.6 years in those with recurrent APO. In models adjusted for sociodemographic characteristics and stroke risk factors, risk of stroke was greater in women with 1 APO (adjusted hazard ratio, 1.3 [95% CI, 1.2–1.4]) and recurrent APO (adjusted hazard ratio, 1.4 [95% CI, 1.2–1.7]) compared with those with no APO. Women with recurrent APO had more than twice the stroke risk before age 45 (adjusted odds ratio, 2.1 [95% CI, 1.5–3.1]) compared with those without APO. CONCLUSIONS: Women who experience APO have earlier onset of cerebrovascular disease, with the earliest onset in those with more than 1 affected pregnancy.

The limited capacity of the brain to retain information in working memory has been well-known and studied for decades, yet the root of this limitation remains unclear. Here we built sensory-cognitive neural network models of working memory that perform tasks using raw visual stimuli. Contrary to intuitions that working memory capacity limitation stems from memory or cognitive constraints, we found that pre-training the sensory region of our models with natural images imposes sufficient constraints on models to exhibit a wide range of human-like behaviors in visual working memory tasks designed to probe capacity. Examining the neural mechanisms in our model reveals that capacity limitation mainly arises in a bottom-up manner. Our models offer a principled and functionally grounded explanation for the working memory capacity limitation without parameter fitting to behavioral data or much hyperparameter tuning. This work highlights the importance of developing models with realistic sensory processing even when investigating memory and other high-level cognitive phenomena.

In the realm of single-cell analysis, computational approaches have brought an increasing number of fantastic prospects for innovation and invention. Meanwhile, it also presents enormous hurdles to reproducing the results of these models due to their diversity and complexity. In addition, the lack of gold-standard benchmark datasets, metrics, and implementations prevents systematic evaluations and fair comparisons of available methods. Thus, we introduce the DANCE platform, the first standard, generic, and extensible benchmark platform for accessing and evaluating computational methods across the spectrum of benchmark datasets for numerous single-cell analysis tasks. Currently, DANCE supports 3 modules and 8 popular tasks with 32 state-of-art methods on 21 benchmark datasets. People can easily reproduce the results of supported algorithms across major benchmark datasets via minimal efforts (e.g., only one command line). In addition, DANCE provides an ecosystem of deep learning architectures and tools for researchers to develop their own models conveniently. The goal of DANCE is to accelerate the development of deep learning models with complete validation and facilitate the overall advancement of single-cell analysis research. DANCE is an open-source python package that welcomes all kinds of contributions. All resources are integrated and available athttps://omicsml.ai/.

Population isolates such as Finland provide benefits in genetic studies because the allelic spectrum of damaging alleles in any gene is often concentrated on a small number of low-frequency variants (0.1% ≤ minor allele frequency < 5%), which survived the founding bottleneck, as opposed to being distributed over a much larger number of ultra--rare variants. While this advantage is well-- established in Mendelian genetics, its value in common disease genetics has been less explored. FinnGen aims to study the genome and national health register data of 500,000 Finns, already reaching 224,737 genotyped and phenotyped participants. Given the relatively high median age of participants (63 years) and dominance of hospital-based recruitment, FinnGen is enriched for many disease endpoints often underrepresented in population-based studies (e.g., rarer immune-mediated diseases and late onset degenerative and ophthalmologic endpoints). We report here a genome-wide association study (GWAS) of 1,932 clinical endpoints defined from nationwide health registries. We identify genome--wide significant associations at 2,491 independent loci. Among these, finemapping implicates 148 putatively causal coding variants associated with 202 endpoints, 104 with low allele frequency (AFWe studied a benchmark set of 15 diseases that had previously been investigated in large genome-wide association studies. FinnGen discovery analyses were meta-analysed in Estonian and UK biobanks. We identify 30 novel associations, primarily low-frequency variants strongly enriched, in or specific to, the Finnish population and Uralic language family neighbors in Estonia and Russia.These findings demonstrate the power of bottlenecked populations to find unique entry points into the biology of common diseases through low-frequency, high impact variants. Such high impact variants have a potential to contribute to medical translation including drug discovery.

The human ability to adaptively implement a wide variety of tasks is thought to emerge from the dynamic transformation of cognitive information. We hypothesized that these transformations are implemented via conjunctive activations in “conjunction hubs”—brain regions that selectively integrate sensory, cognitive, and motor activations. We used recent advances in using functional connectivity to map the flow of activity between brain regions to construct a task-performing neural network model from fMRI data during a cognitive control task. We verified the importance of conjunction hubs in cognitive computations by simulating neural activity flow over this empirically-estimated functional connectivity model. These empirically-specified simulations produced above-chance task performance (motor responses) by integrating sensory and task rule activations in conjunction hubs. These findings reveal the role of conjunction hubs in supporting flexible cognitive computations, while demonstrating the feasibility of using empirically-estimated neural network models to gain insight into cognitive computations in the human brain.

Finding biomarkers that provide shared link between disease severity, drug-induced pharmacodynamic effects and response status in human trials can provide number of values for patient benefits: elucidating current therapeutic mechanism-of-action, and, back-translating to fast-track development of next-generation therapeutics. Both opportunities are predicated on proactive generation of human molecular profiles that capture longitudinal trajectories before and after pharmacological intervention. Here, we present the largest plasma proteomic biomarker dataset available to-date and the corresponding analyses from placebo-controlled Phase III clinical trials of the phosphodiesterase type 4 inhibitor apremilast in psoriasis (PSOR), psoriatic arthritis (PsA), and ankylosing spondylitis (AS) from 526 subjects overall. Using approximately 150 plasma analytes tracked across three time points, we identified IL-17A and KLK-7 as biomarkers for disease severity and apremilast pharmacodynamic effect in psoriasis patients. Combined decline rate of KLK-7, PEDF, MDC and ANGPTL4 by Week 16 represented biomarkers for the responder subgroup, shedding insights into therapeutic mechanisms. In ankylosing spondylitis patients, IL-6 and LRG-1 were identified as biomarkers with concordance to disease severity. Apremilast-induced LRG-1 increase was consistent with the overall lack of efficacy in ankylosing spondylitis. Taken together, these findings expanded the mechanistic knowledge base of apremilast and provided translational foundations to accelerate future efforts including compound differentiation, combination, and repurposing.

Objectives: To compare the outcomes of endoscopic versus microscopic tympanoplasty during the initial period of a surgeon adopting the new endoscopic technique and teaching the surgical approach to residents assisting in surgery. Methods: Retrospective medical chart review of 60 consecutive operations for repair of isolated tympanic membrane perforations from 2011 to 2016 performed by a single surgeon assisted by residents in an academic teaching hospital. The outcomes of 20 ears repaired microscopically before the senior author adopted endoscopic ear surgery (Group A) were compared with the outcomes of the first 20 ears that were attempted with endoscopic surgery (Group B) and the next 20 ears performed endoscopically (Group C). Sixty ear operations were performed on 52 patients as 8 patients had bilateral ear surgery. Results: The tympanic membrane closure rate was 80% for Group A, 80% for Group B, and 95% for Group C. Mean air-bone gap improvement was 12.8 dB in Group A, 8.3 dB in Group B, and 12.1 dB in Group C. Mean duration of surgery was 99.2 minutes in Group A, 91.3 minutes in Group B, and 90.5 minutes in Group C. In Group B, 20% of the ears (4/20) were converted to a microscopic approach; in Group C, none required conversion. Conclusions: Maintenance of good outcomes and similar results can be maintained during a surgeon’s transition to adopting endoscopic tympanoplasty and teaching it to residents.

The brain has the ability to flexibly perform many tasks, but the underlying mechanism cannot be elucidated in traditional experimental and modeling studies designed for one task at a time. Here, we trained single network models to perform 20 cognitive tasks that depend on working memory, decision making, categorization, and inhibitory control. We found that after training, recurrent units can develop into clusters that are functionally specialized for different cognitive processes, and we introduce a simple yet effective measure to quantify relationships between single-unit neural representations of tasks. Learning often gives rise to compositionality of task representations, a critical feature for cognitive flexibility, whereby one task can be performed by recombining instructions for other tasks. Finally, networks developed mixed task selectivity similar to recorded prefrontal neurons after learning multiple tasks sequentially with a continual-learning technique. This work provides a computational platform to investigate neural representations of many cognitive tasks.

SummaryThe convergent evolution of the fly and mouse olfactory system led us to ask whether the anatomic connectivity and functional logic of olfactory circuits would evolve in artificial neural networks trained to perform olfactory tasks. Artificial networks trained to classify odor identity recapitulate the connectivity inherent in the olfactory system. Input units are driven by a single receptor type, and units driven by the same receptor converge to form a glomerulus. Glomeruli exhibit sparse, unstructured connectivity to a larger, expansion layer of Kenyon cells. When trained to both classify odor identity and to impart innate valence onto odors, the network develops independent pathways for identity and valence classification. Thus, the defining features of fly and mouse olfactory systems also evolved in artificial neural networks trained to perform olfactory tasks. This implies that convergent evolution reflects an underlying logic rather than shared developmental principles.

Over the course of a lifetime, a continual stream of information is encoded and retrieved from memory. To explore the synaptic mechanisms that enable this ongoing process, we consider a continual familiarity detection task in which a subject must report whether an image has been previously encountered. We design a class of feedforward neural network models endowed with biologically plausible synaptic plasticity dynamics, the parameters of which are meta-learned to optimize familiarity detection over long delay intervals. After training, we find that anti-Hebbian plasticity leads to better performance than Hebbian and replicates experimental results from the inferotemporal cortex, including repetition suppression. Unlike previous models, this network both operates continuously without requiring any synaptic resets and generalizes to intervals it has not been trained on. We demonstrate this not only for uncorrelated random stimuli but also for images of real-world objects. Our work suggests a biologically plausible mechanism for continual learning, and demonstrates an effective application of machine learning for neuroscience discovery.

The strategies found by animals facing a new task are determined both by individual experience and by structural priors evolved to leverage the statistics of natural environments. Rats can quickly learn to capitalize on the trial sequence correlations of two-alternative forced choice (2AFC) tasks after correct trials, but consistently deviate from optimal behavior after error trials, when they waive the accumulated evidence. To understand this outcome-dependent gating, we first show that Recurrent Neural Networks (RNNs) trained in the same 2AFC task outperform rats as they can readily learn to use across-trial information both after correct and error trials. We hypothesize that, while RNNs can optimize their behavior in the 2AFC task without any a priori restrictions, rats’ strategy is constrained by a structural prior adapted to a natural environment in which rewarded and non-rewarded actions provide largely asymmetric information. When pre-training RNNs in a more ecological task with more than two possible choices, networks develop a strategy by which they gate off the across-trial evidence after errors, mimicking rats’ behavior. Population analyses show that the pre-trained networks form an accurate representation of the sequence statistics independently of the outcome in the previous trial. After error trials, gating is implemented by a change in the network dynamics which temporarily decouples the categorization of the stimulus from the across-trial accumulated evidence. Our results suggest that the suboptimal behavior observed in rats reflects the influence of a structural prior that reacts to errors by isolating the network decision dynamics from the context, ultimately constraining the performance in a 2AFC laboratory task.

The human ability to adaptively implement a wide variety of tasks is thought to emerge from the dynamic transformation of cognitive information. We hypothesized that these transformations are implemented via conjunctive activations in conjunction hubs – brain regions that selectively integrate sensory, cognitive, and motor activations. We used recent advances in using functional connectivity to map the flow of activity between brain regions to construct a task-performing neural network model from fMRI data during a cognitive control task. We verified the importance of conjunction hubs in cognitive computations by simulating neural activity flow over this empirically-estimated functional connectivity model. These empirically-specified simulations produced above-chance task performance (motor responses) by integrating sensory and task rule activations in conjunction hubs. These findings reveal the role of conjunction hubs in supporting flexible cognitive computations, while demonstrating the feasibility of using empirically-estimated neural network models to gain insight into cognitive computations in the human brain.

Summary Over the course of a lifetime, we process a continual stream of information. Extracted from this stream, memories must be efficiently encoded and stored in an addressable manner for retrieval. To explore potential mechanisms, we consider a familiarity detection task in which a subject reports whether an image has been previously encountered. We design a feedforward network endowed with synaptic plasticity and an addressing matrix, meta-learned to optimize familiarity detection over long intervals. We find that anti-Hebbian plasticity leads to better performance than Hebbian plasticity and replicates experimental results such as repetition suppression. A combinatorial addressing function emerges, selecting a unique neuron as an index into the synaptic memory matrix for storage or retrieval. Unlike previous models, this network operates continuously and generalizes to intervals it has not been trained on. Our work suggests a biologically plausible mechanism for continual learning and demonstrates an effective application of machine learning for neuroscience discovery.

Artificial neural networks (ANNs) are essential tools in machine learning that have drawn increasing attention in neuroscience. Besides offering powerful techniques for data analysis, ANNs provide a new approach for neuroscientists to build models for complex behaviors, heterogeneous neural activity and circuit connectivity, as well as to explore optimization in neural systems, in ways that traditional models are not designed for. In this pedagogical Primer, we introduce ANNs and demonstrate how they have been fruitfully deployed to study neuroscientific questions. We first discuss basic concepts and methods of ANNs. Then, with a focus on bringing this mathematical framework closer to neurobiology, we detail how to customize the analysis, structure, and learning of ANNs to better address a wide range of challenges in brain research. To help the readers garner hands-on experience, this Primer is accompanied with tutorial-style code in PyTorch and Jupyter Notebook, covering major topics.

In the mammalian neocortex, an area typically receives inputs from, and projects to, dozens of other areas. Mechanisms are needed to flexibly route information to the right place at the right time, which we term “pathway gating”. For instance, a region in your brain that receives signals from both visual and auditory pathways may want to “gate in” the visual pathway while “gating out” the auditory pathway when you try to read a book surrounded by people in a noisy cafe. In this review, we marshall experimental and computational evidence in support of a circuit mechanism for flexible pathway gating realized by a disinhibitory motif. Moreover, recent work shows an increasing preponderance of this disinhibitory motif from sensory areas to association areas of the mammalian cortex. Pathway input gating is briefly compared with alternative or complementary gating mechanisms. Predictions and open questions for future research on this puzzle about the complex brain system will be discussed.

Most biological and artificial neural systems are capable of completing multiple tasks. However, the neural mechanism by which multiple tasks are accomplished within the same system is largely unclear. We start by discussing how different tasks can be related, and methods to generate large sets of inter-related tasks to study how neural networks and animals perform multiple tasks. We then argue that there are mechanisms that emphasize either specialization or flexibility. We will review two such neural mechanisms underlying multiple tasks at the neuronal level (modularity and mixed selectivity), and discuss how different mechanisms can emerge depending on training methods in neural networks.

One of the most fundamental organizational principles of the brain is the separation of excitatory (E) and inhibitory (I) neurons. In addition to their opposing effects on post-synaptic neurons, E and I cells tend to differ in their selectivity and connectivity. Although many such differences have been characterized experimentally, it is not clear why they exist in the first place. We studied this question in an artificial neural network equipped with multiple E and I cell types. We found that a deep convolutional recurrent network trained to perform an object classification task was able to capture salient distinctions between E and I neurons. We explored the necessary conditions for the network to develop distinct selectivity and connectivity across cell types. We found that neurons that project to higher-order areas will have greater stimulus selectivity, regardless of whether they are excitatory or not. Sparser connectivity is required for higher selectivity, but only when the recurrent connections are excitatory. These findings demonstrate that the differences observed across E and I neurons are not independent, and can be explained using a smaller number of factors.

Background and Aims Advances in cross-sectional imaging have resulted in increased detection of intraductal papillary mucinous neoplasms (IPMNs), and their management remains controversial. At present, there is no reliable noninvasive method to distinguish between indolent and high risk IPMNs. We performed extracellular vesicle (EV) analysis to identify markers of malignancy in an attempt to better stratify these lesions. Methods Using a novel ultrasensitive digital extracellular vesicle screening technique (DEST), we measured putative biomarkers of malignancy (MUC1, MUC2, MUC4, MUC5AC, MUC6, Das-1, STMN1, TSP1, TSP2, EGFR, EpCAM, GPC1, WNT-2, EphA2, S100A4, PSCA, MUC13, ZEB1, PLEC1, HOOK1, PTPN6, and FBN1) in EV from patient-derived cell lines and then on circulating EV obtained from peripheral blood drawn from patients with IPMNs. We enrolled a total of 133 patients in two separate cohorts: a clinical discovery cohort (n = 86) and a validation cohort (n = 47). Results From 16 validated EV proteins in plasma samples collected from the discovery cohort, only MUC5AC showed significantly higher levels in high-grade lesions. Of the 11 patients with invasive IPMN (inv/HG), 9 had high MUC5AC expression in plasma EV of the 11 patients with high-grade dysplasia alone, only 1 had high MUC5AC expression (sensitivity of 82%, specificity of 100%). These findings were corroborated in a separate validation cohort. The addition of MUC5AC as a biomarker to imaging and high-risk stigmata allowed detection of all cases requiring surgery, whereas imaging and high-risk stigmata alone would have missed 5 of 14 cases (36%). Conclusions MUC5AC in circulating EV can predict the presence of invasive carcinoma within IPMN. This approach has the potential to improve the management and follow-up of patients with IPMN including avoiding unnecessary surgery.

Permanent tracheostomy, described first in the 1960s as a treatment for obstructive sleep apnea (OSA), was one of the earliest methods used in the treatment of OSA before the popularization of positive pressure methods in the 1980s. It was reported that tracheostomy in patients with OSA resulted in complete resolution of the systemic and pulmonary hypertension as well as hypersomnia. Studies have shown that tracheostomy remains the only consistently effective method for relieving OSA in treating patients with body mass index >40kg/m 2 . However, given the quality-of-life changes associated with a tracheostomy, it has remained a surgical option of last resort. In this article, we discuss the indications and techniques of permanent tracheostomy.

Background:Apremilast (APR), an oral phosphodiesterase 4 (PDE4) inhibitor, modulates inflammatory mediators1and has demonstrated efficacy in treating oral ulcers in a phase III Behçet’s syndrome study (BCT-002 [RELIEF]).2Objectives:To conduct an exploratory analysis of genetic polymorphisms, plasma biomarkers, and blood leukocytes with clinical response in RELIEF.Methods:Subjects with active Behçet’s disease (BD) were randomized (1:1) to APR 30 mg twice daily or placebo (PBO). The primary clinical efficacy endpoint was the area under the curve for the number of oral ulcers through Week 12 (AUCWk0-12). Among the 207 subjects enrolled, 140 provided consent for DNA genotyping, 116 for plasma biomarker testing, and 96 for leukocyte subset testing. Genotyping was performed on the Illumina Omni2.5 BeadChip (Covance Genomics Laboratory). TNF-α, IL-6, interferon-γ, and IL-17A levels were measured using Simoa Single Molecule Array; IL-8 and IL-23 were measured using the Human DiscoveryMAP multiplex panel (Myriad RBM). Th17, Treg, and CD3 T cells were counted using bisulfite-specific RT-PCR (Epiontis Gmbh). A rank ANCOVA model was used to estimate between-treatment differences (APR vs. PBO) in percent change from baseline for each biomarker/leukocyte subtype over the 12 weeks of treatment. Within each treatment group, the correlation of percent change from baseline at Week 12 in biomarker/leukocyte subtype with the primary efficacy endpoint AUCWk0-12was examined using a univariate regression model. A separate regression model was used to assess the interaction between treatment and the biomarker/leukocyte subtype clinical response.Results:Pharmacogenetic analysis of BD risk variants in HLA-B, IL-10, TLR2, ACE, TNF, GIMAP, PDGFRL, and UBAC2 + 55 genes associated with PDE4 biology yielded no candidate variants that were significantly associated with response to APR or PBO at a Bonferroni-correctedPvalue of 2 x 10−6. Clinical response to APR with respect to HLA-B51 yielded an odds ratio (OR) of 1.21 (95% CI, 0.53-2.75), indicating no significant relationship (Figure 1). Pharmacodynamic changes for IL-6, IL-3, IL-17A, IL-23, and TNF-α were not statistically significant. APR treatment was associated with a significant change in interferon-γ (mean: +107.4%; median: −19.2%) vs. PBO (mean: +78.8%; median: +7.9%) (P=0.0077). Using a univariate regression model, TNF-α showed strong positive correlation with AUCWk0-12in the APR group (r=0.90;P=0.0140); IL-8 had weak positive correlation with AUCWk0-12in the APR group (r=0.04;P=0.0333). A significant negative correlation was observed between the percent change from baseline in the number of Th17 cells and AUCWk0-12in the APR group (r=−0.79;P=0.0392) and a significant positive correlation was observed with the percent change from baseline in the number of Treg cells and AUCWk0-12in the PBO group (r=0.94;P=0.0182). Of all the biomarkers and leukocyte subtypes examined in a regression model using treatment as a factor, only Treg had a statistically significant treatment interaction (P=0.0069).Conclusion:Although there were no genetic predictors of clinical response to APR treatment, strong correlation was observed between the percent change from baseline in plasma TNF-α with AUCWk0-12in the APR group. A negative correlation was observed between percent change from baseline in Th17 cells and AUCWk0-12in the APR group and a positive association was observed between Treg cells and AUCW0-12in the PBO group.References:[1]Schafer P.Biochem Pharmacol. 2012; 83:1583-1590. 2. Hatemi G, et al. Presented at: ACR/ARHP Annual Meeting; November 8–13, 2019; Atlanta, GA. Presentation 0946.Disclosure of Interests: :Joseph Maranville Employee of: Celgene Corporation – employment at the time of study conduct, Irina Medvedeva Employee of: Celgene Corporation – employment at the time of study conduct, Robert Yang Employee of: Celgene Corporation – employment at the time of study conduct, Mindy Chen Employee of: Amgen Inc. – employment; Celgene Corporation – employment at the time of the conduct, Lorraine (Ruoying) Fang Employee of: Celgene Corporation – employment at the time of study conduct, Sandra Collazo Employee of: Amgen Inc. – employment; Celgene Corporation – employment at the time of the conduct, Shannon McCue Employee of: Amgen Inc. – employment; Celgene Corporation – employment at the time of the conduct, Peter Schafer Employee of: Bristol-Myers Squibb – employment; Celgene Corporation – employment at the time of study conduct