Search

Background Identifying nuclease-induced double-stranded breaks in DNA on a genome-wide scale is critical for assessing the safety and efficacy of genome editing therapies. We previously demonstrated that after administering adeno-associated viral (AAV) vector-mediated genome-editing strategies in vivo, vector sequences integrated into the host organism’s genomic DNA at double-stranded breaks. Thus, identifying the genomic location of inserted AAV sequences would enable us to identify DSB events, mainly derived from the nuclease on- and off-target activity. Results Here, we developed a next-generation sequencing assay that detects insertions of specific AAV vector sequences called inverted terminal repeats (ITRs). This assay, ITR-Seq, enables us to identify off-target nuclease activity in vivo. Using ITR-Seq, we analyzed liver DNA samples of rhesus macaques treated with AAV vectors expressing a meganuclease. We found dose-dependent off-target activity and reductions in off-target events induced by further meganuclease development. In mice, we identified the genomic locations of ITR integration after treatment with Cas9 nucleases and their corresponding single-guide RNAs. Conclusions In sum, ITR-Seq is a powerful method for identifying off-target sequences induced by AAV vector-delivered genome-editing nucleases. ITR-Seq will help us understand the specificity and efficacy of different genome-editing nucleases in animal models and clinical studies. This information can help enhance the safety profile of gene-editing therapies.

Given the ubiquitous nature of numbers in text, reasoning with numbers to perform simple calculations is an important skill of AI systems. While many datasets and models have been developed to this end, state-of-the-art AI systems are brittle; failing to perform the underlying mathematical reasoning when they appear in a slightly different scenario. Drawing inspiration from GLUE that was proposed in the context of natural language understanding, we propose NumGLUE, a multi-task benchmark that evaluates the performance of AI systems on eight different tasks, that at their core require simple arithmetic understanding. We show that this benchmark is far from being solved with neural models including state-of-the-art large-scale language models performing significantly worse than humans (lower by 46.4%). Further, NumGLUE promotes sharing knowledge across tasks, especially those with limited training data as evidenced by the superior performance (average gain of 3.4% on each task) when a model is jointly trained on all the tasks as opposed to task-specific modeling. Finally, we hope that NumGLUE will encourage systems that perform robust and general arithmetic reasoning within language, a first step towards being able to perform more complex mathematical reasoning., ACL 2022

Precise specification of the research question and associated treatment effect of interest is essential in clinical research, yet recent work shows that they are often incompletely specified. The ICH E9 (R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials introduces a framework that supports researchers in precisely and transparently specifying the treatment effect they aim to estimate in their clinical trial. In this paper, we present practical examples to demonstrate to all researchers involved in clinical trials how estimands can help them to specify the research question, lead to a better understanding of the treatment effect to be estimated and hence increase the probability of success of the trial.

Defeasible reasoning is the mode of reasoning where conclusions can be overturned by taking into account new evidence. Existing cognitive science literature on defeasible reasoning suggests that a person forms a mental model of the problem scenario before answering questions. Our research goal asks whether neural models can similarly benefit from envisioning the question scenario before answering a defeasible query. Our approach is, given a question, to have a model first create a graph of relevant influences, and then leverage that graph as an additional input when answering the question. Our system, CURIOUS, achieves a new state-of-the-art on three different defeasible reasoning datasets. This result is significant as it illustrates that performance can be improved by guiding a system to "think about" a question and explicitly model the scenario, rather than answering reflexively. Code, data, and pre-trained models are located at https://github.com/madaan/thinkaboutit., EMNLP 2021

Experimental procedures have been developed to provide at the laboratory scale an efficient and relatively inexpensive continuous flow processing system to undertake metathesis chemistry. Glass discs, coated with porous silica particles impregnated with Grubbs (I) generation or Hoveyda–Grubbs (II) generation catalysts through non-covalent interactions, have been successfully used as annular zones for in situ metathesis reactions. The supported catalysts were active for ring closing metathesis and self-metathesis reactions with a variety of substrates with nearly quantitative conversion depending on the substrate structure. Advantages of these procedures include the ability to conduct reactions efficiently at small (low mmole) scale under ambient laboratory conditions, thereby avoiding the need for high energy/thermal procedures. Multiple experiments can be undertaken during the same day if catalyst screening is required. Moreover, separation of starting materials from products can be achieved using the same unit operations, avoiding the need for additional purification steps, thereby reducing the total solvent consumption.

A procedure for producing a climatology of tropical wind shear from climate-model output is presented. The procedure is designed to find grid columns in the model where the organization of convection may be present. The climate-model output consists of east–west and north–south wind profiles at 20 equally spaced pressure levels from 1000 hPa to 50 hPa, and the Convective Available Potential Energy (CAPE) as diagnosed by the model’s Convection Parametrization Scheme (CPS). The procedure begins by filtering the wind profiles based on their maximum shear, and on a CAPE threshold of 100 J kg−1. The filtered profiles are normalized using the maximum wind speed at each pressure level, and rotated to align the wind at 850 hPa. From each of the filtered profiles, a sample has been produced with 40 dimensions (20 for each wind direction). The number of dimensions is reduced by using Principal Component Analysis (PCA), where the requirement is that 90 % of the variance must be explained by the principal components. This requires keeping the first seven leading principal components. The samples, as represented by their principal components, can then be clustered using the K-Means Clustering Algorithm (KMCA). 10 clusters are chosen to represent the samples, and the median of each cluster defines a Representative Wind Profile (RWP) – a profile that represents the shear conditions of the wind profiles produced by the climate model. The RWPs are analysed, first in terms of their vertical structure, and then in terms of their geographical and temporal distributions. We find that the RWPs have some features often associated with the organization of convection, such as low-level and mid-level shear. Some of the RWPs can be matched with wind profiles taken from case studies of organization of convection, such as squall lines seen in Tropical Ocean Global Atmosphere, Coupled Atmosphere Ocean Research Experiment (TOGA–COARE). The RWPs’ geographical distributions show that each RWP occurs preferentially in certain regions. Six of the RWPs occur preferentially over land, while three occur preferentially over oceans. The temporal distribution of RWPs shows that they occur preferentially at certain times of the year, with the distributions having mainly one or two modes. Their geographical and temporal distributions are compared with those seen in previous studies of organized convection, and some broad and specific similarities are noted. By performing the analysis on climate-model output, we lay the foundations for the development of the representation of shear-induced organization in a CPS. This would use the same methodology to diagnose where the organization of convection occurs, and modify the CPS in an appropriate manner to represent it.

We present the first dataset for tracking state changes in procedural text from arbitrary domains by using an unrestricted (open) vocabulary. For example, in a text describing fog removal using potatoes, a car window may transition between being foggy, sticky,opaque, and clear. Previous formulations of this task provide the text and entities involved,and ask how those entities change for just a small, pre-defined set of attributes (e.g., location), limiting their fidelity. Our solution is a new task formulation where given just a procedural text as input, the task is to generate a set of state change tuples(entity, at-tribute, before-state, after-state)for each step,where the entity, attribute, and state values must be predicted from an open vocabulary. Using crowdsourcing, we create OPENPI1, a high-quality (91.5% coverage as judged by humans and completely vetted), and large-scale dataset comprising 29,928 state changes over 4,050 sentences from 810 procedural real-world paragraphs from WikiHow.com. A current state-of-the-art generation model on this task achieves 16.1% F1 based on BLEU metric, leaving enough room for novel model architectures., To appear in EMNLP 2020

We propose a general framework called Text Modular Networks(TMNs) for building interpretable systems that learn to solve complex tasks by decomposing them into simpler ones solvable by existing models. To ensure solvability of simpler tasks, TMNs learn the textual input-output behavior (i.e., language) of existing models through their datasets. This differs from prior decomposition-based approaches which, besides being designed specifically for each complex task, produce decompositions independent of existing sub-models. Specifically, we focus on Question Answering (QA) and show how to train a next-question generator to sequentially produce sub-questions targeting appropriate sub-models, without additional human annotation. These sub-questions and answers provide a faithful natural language explanation of the model's reasoning. We use this framework to build ModularQA, a system that can answer multi-hop reasoning questions by decomposing them into sub-questions answerable by a neural factoid single-span QA model and a symbolic calculator. Our experiments show that ModularQA is more versatile than existing explainable systems for DROP and HotpotQA datasets, is more robust than state-of-the-art blackbox (uninterpretable) systems, and generates more understandable and trustworthy explanations compared to prior work., Accepted to NAACL 2021

Question answering (QA) tasks have been posed using a variety of formats, such as extractive span selection, multiple choice, etc. This has led to format-specialized models, and even to an implicit division in the QA community. We argue that such boundaries are artificial and perhaps unnecessary, given the reasoning abilities we seek to teach are not governed by the format. As evidence, we use the latest advances in language modeling to build a single pre-trained QA model, UnifiedQA, that performs surprisingly well across 17 QA datasets spanning 4 diverse formats. UnifiedQA performs on par with 9 different models that were trained on individual datasets themselves. Even when faced with 12 unseen datasets of observed formats, UnifiedQA performs surprisingly well, showing strong generalization from its out-of-format training data. Finally, simply fine-tuning this pre-trained QA model into specialized models results in a new state of the art on 6 datasets, establishing UnifiedQA as a strong starting point for building QA systems., EMNLP 2020 (Findings)

Beginning with McCarthy's Advice Taker (1959), AI has pursued the goal of providing a system with explicit, general knowledge and having the system reason over that knowledge. However, expressing the knowledge in a formal (logical or probabilistic) representation has been a major obstacle to this research. This paper investigates a modern approach to this problem where the facts and rules are provided as natural language sentences, thus bypassing a formal representation. We train transformers to reason (or emulate reasoning) over these sentences using synthetically generated data. Our models, that we call RuleTakers, provide the first empirical demonstration that this kind of soft reasoning over language is learnable, can achieve high (99%) accuracy, and generalizes to test data requiring substantially deeper chaining than seen during training (95%+ scores). We also demonstrate that the models transfer well to two hand-authored rulebases, and to rulebases paraphrased into more natural language. These findings are significant as it suggests a new role for transformers, namely as limited "soft theorem provers" operating over explicit theories in language. This in turn suggests new possibilities for explainability, correctability, and counterfactual reasoning in question-answering., IJCAI 2020

In this study, bulk mass flux formulations for turbulent fluxes are evaluated for shallow and deep convection using large-eddy simulation data. The bulk mass flux approximation neglects two sources of variability: the interobject variability due to differences between the average properties of different cloud objects, and the intraobject variability due to perturbations within each cloud object. Using a simple cloud–environment decomposition, the interobject and intraobject contributions to the heat flux are comparable in magnitude with that from the bulk mass flux approximation, but do not share a similar vertical distribution, and so cannot be parameterized with a rescaling method. A downgradient assumption is also not appropriate to parameterize the neglected flux contributions because a nonnegligible part is associated with nonlocal buoyant structures. A spectral analysis further suggests the presence of fine structures within the clouds. These points motivate investigations in which the vertical transports are decomposed based on the distribution of vertical velocity. As a result, a “core-cloak” conceptual model is proposed to improve the representation of total vertical fluxes, composed of a strong and a weak draft for both the updrafts and downdrafts. It is shown that the core-cloak representation can well capture the magnitude and vertical distribution of heat and moisture fluxes for both shallow and deep convection.

We introduce the first open-domain dataset, called QuaRTz, for reasoning about textual qualitative relationships. QuaRTz contains general qualitative statements, e.g., "A sunscreen with a higher SPF protects the skin longer.", twinned with 3864 crowdsourced situated questions, e.g., "Billy is wearing sunscreen with a lower SPF than Lucy. Who will be best protected from the sun?", plus annotations of the properties being compared. Unlike previous datasets, the general knowledge is textual and not tied to a fixed set of relationships, and tests a system's ability to comprehend and apply textual qualitative knowledge in a novel setting. We find state-of-the-art results are substantially (20%) below human performance, presenting an open challenge to the NLP community., EMNLP'19

While in recent years machine learning (ML) based approaches have been the popular approach in developing endto-end question answering systems, such systems often struggle when additional knowledge is needed to correctly answer the questions. Proposed alternatives involve translating the question and the natural language text to a logical representation and then use logical reasoning. However, this alternative falters when the size of the text gets bigger. To address this we propose an approach that does logical reasoning over premises written in natural language text. The proposed method uses recent features of Answer Set Programming (ASP) to call external NLP modules (which may be based on ML) which perform simple textual entailment. To test our approach we develop a corpus based on the life cycle questions and showed that Our system achieves up to 18% performance gain when compared to standard MCQ solvers.

International audience; The FRANC project (Forecasting Rainfall exploiting new data Assimilation techniques and Novel observations of Convection) has researched improvements in numerical weather prediction of convective rainfall via the reduction of initial condition uncertainty. This article provides an overview of the project's achievements. We highlight new radar techniques: correcting for attenuation of the radar return; correction for beams that are over 90% blocked by trees or towers close to the radar; and direct assimilation of radar reflectivity and refractivity. We discuss the treatment of uncertainty in data assimilation: new methods for estimation of observation uncertainties with novel applications to Doppler radar winds, Atmospheric Motion Vectors, and satellite radiances; a new algorithm for implementation of spatially-correlated observation error statistics in operational data assimilation; and innovative treatment of moist processes in the background error covariance model. We present results indicating a link between the spatial predictability of convection and convective regimes, with potential to allow improved forecast interpretation. The research was carried out as a partnership between University researchers and the Met Office (UK). We discuss the benefits of this approach and the impact of our research, which has helped to improve operational forecasts for convective rainfall events.

Many natural language questions require recognizing and reasoning with qualitative relationships (e.g., in science, economics, and medicine), but are challenging to answer with corpus-based methods. Qualitative modeling provides tools that support such reasoning, but the semantic parsing task of mapping questions into those models has formidable challenges. We present QuaRel, a dataset of diverse story questions involving qualitative relationships that characterize these challenges, and techniques that begin to address them. The dataset has 2771 questions relating 19 different types of quantities. For example, "Jenny observes that the robot vacuum cleaner moves slower on the living room carpet than on the bedroom carpet. Which carpet has more friction?" We contribute (1) a simple and flexible conceptual framework for representing these kinds of questions; (2) the QuaRel dataset, including logical forms, exemplifying the parsing challenges; and (3) two novel models for this task, built as extensions of type-constrained semantic parsing. The first of these models (called QuaSP+) significantly outperforms off-the-shelf tools on QuaRel. The second (QuaSP+Zero) demonstrates zero-shot capability, i.e., the ability to handle new qualitative relationships without requiring additional training data, something not possible with previous models. This work thus makes inroads into answering complex, qualitative questions that require reasoning, and scaling to new relationships at low cost. The dataset and models are available at http://data.allenai.org/quarel., 9 pages, AAAI 2019

We present a new kind of question answering dataset, OpenBookQA, modeled after open book exams for assessing human understanding of a subject. The open book that comes with our questions is a set of 1329 elementary level science facts. Roughly 6000 questions probe an understanding of these facts and their application to novel situations. This requires combining an open book fact (e.g., metals conduct electricity) with broad common knowledge (e.g., a suit of armor is made of metal) obtained from other sources. While existing QA datasets over documents or knowledge bases, being generally self-contained, focus on linguistic understanding, OpenBookQA probes a deeper understanding of both the topic---in the context of common knowledge---and the language it is expressed in. Human performance on OpenBookQA is close to 92%, but many state-of-the-art pre-trained QA methods perform surprisingly poorly, worse than several simple neural baselines we develop. Our oracle experiments designed to circumvent the knowledge retrieval bottleneck demonstrate the value of both the open book and additional facts. We leave it as a challenge to solve the retrieval problem in this multi-hop setting and to close the large gap to human performance., Published as conference long paper at EMNLP 2018