Interactive and Visual Prompt Engineering for Ad-hoc Task Adaptation with Large Language Models

State-of-the-art neural language models can now be used to solve ad-hoc language tasks through zero-shot prompting without the need for supervised training. This approach has gained popularity in recent years, and researchers have demonstrated prompts that achieve strong accuracy on specific NLP tasks. However, finding a prompt for new tasks requires experimentation. Different prompt templates with different wording choices lead to significant accuracy differences. PromptIDE allows users to experiment with prompt variations, visualize prompt performance, and iteratively optimize prompts. We developed a workflow that allows users to first focus on model feedback using small data before moving on to a large data regime that allows empirical grounding of promising prompts using quantitative measures of the task. The tool then allows easy deployment of the newly created ad-hoc models. We demonstrate the utility of PromptIDE (demo: http://prompt.vizhub.ai) and our workflow using several real-world use cases.

Insulin action in the brain: cell types, circuits, and diseases

Since its discovery over 100 years ago, insulin has been recognized as a key hormone in control of glucose homeostasis. Deficiencies of insulin signaling are central to diabetes and many other disorders. The brain is among the targets of insulin action, and insulin resistance is a major contributor to many diseases, including brain disorders. Here, we summarize key roles of insulin action in the brain and how this involves different brain cell types. Disordered brain insulin signaling can also contribute to neuropsychiatric diseases, affecting brain circuits involved in mood and cognition. Understanding of insulin signaling in different brain cell types/circuits and how these are altered in disease may lead to the development of new therapeutic approaches to these challenging disorders.

Operational response simulation tool for epidemics within refugee and IDP settlements: A scenario-based case study of the Cox's Bazar settlement

The spread of infectious diseases such as COVID-19 presents many challenges to healthcare systems and infrastructures across the world, exacerbating inequalities and leaving the world's most vulnerable populations most affected. Given their density and available infrastructure, refugee and internally displaced person (IDP) settlements can be particularly susceptible to disease spread. In this paper we present an agent-based modeling approach to simulating the spread of disease in refugee and IDP settlements under various non-pharmaceutical intervention strategies. The model, based on the June open-source framework, is informed by data on geography, demographics, comorbidities, physical infrastructure and other parameters obtained from real-world observations and previous literature. The development and testing of this approach focuses on the Cox's Bazar refugee settlement in Bangladesh, although our model is designed to be generalizable to other informal settings. Our findings suggest the encouraging self-isolation at home of mild to severe symptomatic patients, as opposed to the isolation of all positive cases in purpose-built isolation and treatment centers, does not increase the risk of secondary infection meaning the centers can be used to provide hospital support to the most intense cases of COVID-19. Secondly we find that mask wearing in all indoor communal areas can be effective at dampening viral spread, even with low mask efficacy and compliance rates. Finally, we model the effects of reopening learning centers in the settlement under various mitigation strategies. For example, a combination of mask wearing in the classroom, halving attendance regularity to enable physical distancing, and better ventilation can almost completely mitigate the increased risk of infection which keeping the learning centers open may cause. These modeling efforts are being incorporated into decision making processes to inform future planning, and further exercises should be carried out in similar geographies to help protect those most vulnerable.

Extraction of organic chemistry grammar from unsupervised learning of chemical reactions

Humans use different domain languages to represent, explore, and communicate scientific concepts. During the last few hundred years, chemists compiled the language of chemical synthesis inferring a series of "reaction rules" from knowing how atoms rearrange during a chemical transformation, a process called atom-mapping. Atom-mapping is a laborious experimental task and, when tackled with computational methods, requires continuous annotation of chemical reactions and the extension of logically consistent directives. Here, we demonstrate that Transformer Neural Networks learn atom-mapping information between products and reactants without supervision or human labeling. Using the Transformer attention weights, we build a chemically agnostic, attention-guided reaction mapper and extract coherent chemical grammar from unannotated sets of reactions. Our method shows remarkable performance in terms of accuracy and speed, even for strongly imbalanced and chemically complex reactions with nontrivial atom-mapping. It provides the missing link between data-driven and rule-based approaches for numerous chemical reaction tasks.

Managing the SOS Response for Enhanced CRISPR-Cas-Based Recombineering in E. coli through Transient Inhibition of Host RecA Activity

Phage-derived "recombineering" methods are utilized for bacterial genome editing. Recombineering results in a heterogeneous population of modified and unmodified chromosomes, and therefore selection methods, such as CRISPR-Cas9, are required to select for edited clones. Cells can evade CRISPR-Cas-induced cell death through recA-mediated induction of the SOS response. The SOS response increases RecA dependent repair as well as mutation rates through induction of the umuDC error prone polymerase. As a result, CRISPR-Cas selection is more efficient in recA mutants. We report an approach to inhibiting the SOS response and RecA activity through the expression of a mutant dominant negative form of RecA, which incorporates into wild type RecA filaments and inhibits activity. Using a plasmid-based system in which Cas9 and recA mutants are coexpressed, we can achieve increased efficiency and consistency of CRISPR-Cas9-mediated selection and recombineering in E. coli, while reducing the induction of the SOS response. To date, this approach has been shown to be independent of recA genotype and host strain lineage. Using this system, we demonstrate increased CRISPR-Cas selection efficacy with over 10 000 guides covering the E. coli chromosome. The use of dominant negative RecA or homologues may be of broad use in bacterial CRISPR-Cas-based genome editing where the SOS pathways are present.

Reductive methylation and mutation of an anthrax toxin fusion protein modulates its stability and cytotoxicity

We characterized an anti-cancer fusion protein consisting of anthrax lethal factor (LF) and the catalytic domain of Pseudomonas exotoxin A by (i) mutating the N-terminal amino acids and by (ii) reductive methylation to dimethylate all lysines. Dimethylation of lysines was achieved quantitatively and specifically without affecting binding of the fusion protein to PA or decreasing the enzymatic activity of the catalytic moiety. Ubiquitination in vitro was drastically decreased for both the N-terminally mutated and dimethylated variants, and both appeared to be slightly more stable in the cytosol of treated cells. The dimethylated variant showed greatly reduced neutralization by antibodies to LF. The two described modifications offer unique advantages such as increased cytotoxic activity and diminished antibody recognition, and thus may be applicable to other therapeutic proteins that act in the cytosol of cells.

RNA toxicity from the ALS/FTD C9ORF72 expansion is mitigated by antisense intervention

A hexanucleotide GGGGCC repeat expansion in the noncoding region of the C9ORF72 gene is the most common genetic abnormality in familial and sporadic amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). The function of the C9ORF72 protein is unknown, as is the mechanism by which the repeat expansion could cause disease. Induced pluripotent stem cell (iPSC)-differentiated neurons from C9ORF72 ALS patients revealed disease-specific (1) intranuclear GGGGCCexp RNA foci, (2) dysregulated gene expression, (3) sequestration of GGGGCCexp RNA binding protein ADARB2, and (4) susceptibility to excitotoxicity. These pathological and pathogenic characteristics were confirmed in ALS brain and were mitigated with antisense oligonucleotide (ASO) therapeutics to the C9ORF72 transcript or repeat expansion despite the presence of repeat-associated non-ATG translation (RAN) products. These data indicate a toxic RNA gain-of-function mechanism as a cause of C9ORF72 ALS and provide candidate antisense therapeutics and candidate human pharmacodynamic markers for therapy.

Increased expression of glutamate transporter GLT-1 in peritumoral tissue associated with prolonged survival and decreases in tumor growth in a rat model of experimental malignant glioma

OBJECT

Gliomas are known to release excessive amounts of glutamate, inducing glutamate excitotoxic cell death in the peritumoral region and allowing the tumor to grow and to expand. Glutamate transporter upregulation has been shown to be neuroprotective by removing extracellular glutamate in a number of preclinical animal models of neurodegenerative diseases, including amyotrophic lateral sclerosis and Parkinson disease as well as psychiatric disorders such as depression. The authors therefore hypothesized that the protective mechanism of glutamate transporter upregulation would be useful for the treatment of gliomas as well.

METHODS

In this study 9L gliosarcoma cells were treated with a glutamate transporter upregulating agent, thiamphenicol, an antibiotic approved in Europe, which has been shown previously to increase glutamate transporter expression and has recently been validated in a human Phase I biomarker trial for glutamate transporter upregulation. Cells were monitored in vitro for glutamate transporter levels and cell proliferation. In vivo, rats were injected intracranially with 9L cells and were treated with increasing doses of thiamphenicol. Animals were monitored for survival. In addition, postmortem brain tissue was analyzed for tumor size, glutamate transporter levels, and neuron count.

RESULTS

Thiamphenicol showed little effects on proliferation of 9L gliosarcoma cells in vitro and did not change glutamate transporter levels in these cells. However, when delivered locally in an experimental glioma model in rats, thiamphenicol dose dependently (10-5000 μM) significantly increased survival up to 7 days and concomitantly decreased tumor size from 46.2 mm(2) to 10.2 mm(2) when compared with lesions in nontreated controls. Furthermore, immunohistochemical and biochemical analysis of peritumoral tissue confirmed an 84% increase in levels of glutamate transporter protein and a 72% increase in the number of neuronal cells in the tissue adjacent to the tumor.

CONCLUSIONS

These results show that increasing glutamate transporter expression in peritumoral tissue is neuroprotective. It suggests that glutamate transporter upregulation for the treatment of gliomas should be further investigated and potentially be part of a combination therapy with standard chemotherapeutic agents.

Some issues relevant to establishing a universal newborn hearing screening program

This article describes some of the factors relevant to the establishment of a universal newborn hearing screening (UNHS) program. First, the difficulty in providing precise estimates of test sensitivity and specificity are reviewed. This section is followed by hypothetical estimates of overall programmatic costs, first for a fixed number of babies to be screened and then as a function of the number of babies to be screened in a year. Included in these estimates are the costs for equipment, disposables, personnel, and follow-up testing. These estimates are provided for three different screening protocols: auditory brainstem response (ABR) alone, otoacoustic emission (OAE) alone, and OAE followed by ABR only for those babies who failed the OAE screening. If follow-up costs are not included, it is less expensive to screen newborns with OAEs compared with the other two protocols. However, once follow-up testing is included as part of the program costs and there are at least 400 births per year, procedures in which OAEs are performed first, followed by an ABR on those infants who do not pass the OAE test, result in the lowest costs. Hospitals with as few as 400 births per year should expect per-baby costs not exceeding $30, regardless of which protocol is used. For all three protocols, the unit costs decrease as the number of babies screened increases. The final section describes data from a local UNHS program in which all infants are screened first with an OAE test, followed by an ABR test on infants not passing the OAE screening. Idiosyncratic features to this program are described, including the fact that all screening tests are performed by audiologists, who are paid on a part-time basis, adding cost to the program. Even under these circumstances, the unit cost is under $30. These data lead us to conclude that all infants can be screened in a cost-effective manner.

Subjective effects of peak clipping and compression limiting in normal and hearing-impaired children and adults

Despite many advances in hearing-aid signal processing, compression limiting and peak clipping are still used. To date, perceptual studies have been conducted only with adults. The current study was designed to investigate the clarity of peak-clipped and compressed speech for both adults and children. Subjects were 30 normal-hearing and 30 hearing-impaired individuals in three age ranges (7-9, 10-12, and 16-50 years). Stimuli were processed at 60, 70, 75, and 80 dB SPL using peak clipping and at 80 dB SPL using compression limiting. Paired-comparison measures were used to assess the clarity of sentences, and a signal-to-distortion ratio (SDR) based on a measure of coherence between input and output was computed for each condition. For the peak-clipping conditions, there was a decrease in perceived clarity as the input increased from 60 to 80 dB SPL. This perceptual continuum was most apparent for the normal-hearing adults. The normal-hearing 10-12 year olds and the hearing-impaired adults showed a similar, but less pronounced, pattern. In contrast, the remaining three subject groups showed minimal differences in perceived clarity across conditions. Surprisingly, only the two oldest normal-hearing groups showed a clear preference for compression limiting over peak clipping at the highest input level, and only their results were consistent with the pattern of coherence across stimuli. Judgments of clarity by the normal-hearing subjects correlated best with the SDR in the 500-2000-Hz range, while clarity judgments of the hearing-impaired subjects correlated best with the SDR below 1000 Hz.

From laboratory to clinic: a large scale study of distortion product otoacoustic emissions in ears with normal hearing and ears with hearing loss

OBJECTIVES

1) To describe distortion product otoacoustic emission (DPOAE) measurements in large groups of subjects with normal hearing and with hearing loss, and to use these data to provide comprehensive descriptions of DPOAE test performance. 2) To describe the effects of primary frequency and audiometric threshold on the extent to which DPOAE measurements accurately identify auditory status. 3) To develop an approach that describes the probability that any measured response is coming from either a normal or an impaired ear. 4) To develop an approach for representing DPOAE data clinically. 5) To explore the relation between magnitude of hearing loss and DPOAE measurements.

DESIGN

DPOAE measurements were made in 1267 ears of 806 subjects, using stimulus conditions that previously had been demonstrated to result in the greatest separation between normal and impaired ears (i.e., primary levels of 65/55 dB SPL for f1/f2; Stover et al., 1996). Subjects were recruited from local clinical populations and through local advertisements. All data were analyzed using clinical decision theory, including relative operating characteristic (ROC) curves and estimates of areas under these curves (Az). In addition, cumulative distributions were constructed of response properties from both normal and hearing-impaired ears. These cumulative distributions were used to select specific probabilities that measured responses were coming from either the normal or impaired distributions, and to develop an approach for describing clinical DPOAE data.

RESULTS

For no conditions were the distributions of DPOAE responses from normal and impaired ears completely separated, meaning that optimal criterion values would still result in errors in identification of auditory status. Test performance, defined by Az, was best for mid and high frequencies and poorest for lower frequencies and for the highest frequency tested (8000 Hz). Performance was best when normal hearing was defined as audiometric thresholds between 20 and 30 dB HL, with poorer performance for more stringent or lax audiometric criteria.

CONCLUSIONS

Within the limits related to the effects of primary frequency and audiometric criterion, it appears that DPOAE measurements can be used to accurately identify auditory status. An approach is described, using the present data set, that allows one to assign to any measured DPOAE value (DPOAE amplitudes, DPOAE/noise) the probability that the response is coming either from the distribution of normal or impaired responses. In addition, DPOAE/noise systematically decreases as hearing loss increases over the range of hearing losses from 0 to about 40 to 60 dB HL (depending on frequency), thus potentially enabling one to differentiate hearing losses over this range. For hearing losses greater than 50 to 60 dB HL, ears do not produce measurable DPOAEs and thus, no predictive relationship exists.

Time-gated ensemble-averaged imaging through highly scattering media

A previously described ensemble-averaged imaging method [Opt. Lett. 21, 1691 (1996)] is extended by its combination with holographically implemented time-gated imaging. This combined method is shown to extend the effectiveness of the ensemble-averaged method by permitting imaging through thicker diffusers. Experimental results are presented.

A model for integrating technology into a multi-agency community service delivery system

This article presents the results of a 4-year project to develop and test a model for integrating technology into a multi-agency community human services delivery system. "Technology" in this sense refers to both information technology and assistive technology. The project focused on systems change rather than on the creation of comprehensive assistive technology services. Several strategies of the model were developed and tested by the project. The project developed an information/communication tier that lay on top of the existing service delivery system. The information/communication tier was provided to agencies through an electronic network and searchable database, assistive technology advice-giving software, a public awareness campaign, and professional training. The results of these strategies are used to illustrate the Community Assistive Technology Services Network (CATSN) model. The model consists of five modules: 1) client assessment and technology examination, 2) professional and client training and technical assistance, 3) information and referral and needs analysis, 4) special projects, and 5) central coordination and facilitation--the hub. This model is important given recent nationwide initiatives for using assistive technology to provide people with disabilities opportunities for maximum independence, productivity, and integration.