Search

Data-driven scientific discovery requires the iterative integration of scientific domain knowledge, statistical expertise, and an understanding of data semantics to make nuanced analytical decisions, e.g., about which variables, transformations, and statistical models to consider. LM-based agents equipped with planning, memory, and code execution capabilities have the potential to support data-driven science. However, evaluating agents on such open-ended tasks is challenging due to multiple valid approaches, partially correct steps, and different ways to express the same decisions. To address these challenges, we present BLADE, a benchmark to automatically evaluate agents' multifaceted approaches to open-ended research questions. BLADE consists of 12 datasets and research questions drawn from existing scientific literature, with ground truth collected from independent analyses by expert data scientists and researchers. To automatically evaluate agent responses, we developed corresponding computational methods to match different representations of analyses to this ground truth. Though language models possess considerable world knowledge, our evaluation shows that they are often limited to basic analyses. However, agents capable of interacting with the underlying data demonstrate improved, but still non-optimal, diversity in their analytical decision making. Our work enables the evaluation of agents for data-driven science and provides researchers deeper insights into agents' analysis approaches.

Existing statistical methods in causal inference often rely on the assumption that every individual has some chance of receiving any treatment level regardless of its associated covariates, which is known as the positivity condition. This assumption could be violated in observational studies with continuous treatments. In this paper, we present a novel integral estimator of the causal effects with continuous treatments (i.e., dose-response curves) without requiring the positivity condition. Our approach involves estimating the derivative function of the treatment effect on each observed data sample and integrating it to the treatment level of interest so as to address the bias resulting from the lack of positivity condition. The validity of our approach relies on an alternative weaker assumption that can be satisfied by additive confounding models. We provide a fast and reliable numerical recipe for computing our estimator in practice and derive its related asymptotic theory. To conduct valid inference on the dose-response curve and its derivative, we propose using the nonparametric bootstrap and establish its consistency. The practical performances of our proposed estimators are validated through simulation studies and an analysis of the effect of air pollution exposure (PM$_{2.5}$) on cardiovascular mortality rates., Comment: 74 pages (23 pages for the main paper), 4 figures

Molecule-and-text cross-modal representation learning has emerged as a promising direction for enhancing the quality of molecular representation, thereby improving performance in various scientific fields, including drug discovery and materials science. Existing studies adopt a global alignment approach to learn the knowledge from different modalities. These global alignment approaches fail to capture fine-grained information, such as molecular fragments and their corresponding textual description, which is crucial for downstream tasks. Furthermore, it is incapable to model such information using a similar global alignment strategy due to data scarcity of paired local part annotated data from existing datasets. In this paper, we propose Atomas, a multi-modal molecular representation learning framework to jointly learn representations from SMILES string and text. We design a Hierarchical Adaptive Alignment model to concurrently learn the fine-grained fragment correspondence between two modalities and align these representations of fragments in three levels. Additionally, Atomas's end-to-end training framework incorporates the tasks of understanding and generating molecule, thereby supporting a wider range of downstream tasks. In the retrieval task, Atomas exhibits robust generalization ability and outperforms the baseline by 30.8% of recall@1 on average. In the generation task, Atomas achieves state-of-the-art results in both molecule captioning task and molecule generation task. Moreover, the visualization of the Hierarchical Adaptive Alignment model further confirms the chemical significance of our approach. Our codes can be found at https://anonymous.4open.science/r/Atomas-03C3.

The core challenge of de novo protein design lies in creating proteins with specific functions or properties, guided by certain conditions. Current models explore to generate protein using structural and evolutionary guidance, which only provide indirect conditions concerning functions and properties. However, textual annotations of proteins, especially the annotations for protein domains, which directly describe the protein's high-level functionalities, properties, and their correlation with target amino acid sequences, remain unexplored in the context of protein design tasks. In this paper, we propose Protein-Annotation Alignment Generation (PAAG), a multi-modality protein design framework that integrates the textual annotations extracted from protein database for controllable generation in sequence space. Specifically, within a multi-level alignment module, PAAG can explicitly generate proteins containing specific domains conditioned on the corresponding domain annotations, and can even design novel proteins with flexible combinations of different kinds of annotations. Our experimental results underscore the superiority of the aligned protein representations from PAAG over 7 prediction tasks. Furthermore, PAAG demonstrates a nearly sixfold increase in generation success rate (24.7% vs 4.7% in zinc finger, and 54.3% vs 8.7% in the immunoglobulin domain) in comparison to the existing model.

This paper is concerned with inference on the regression function of a high-dimensional linear model when outcomes are missing at random. We propose an estimator which combines a Lasso pilot estimate of the regression function with a bias correction term based on the weighted residuals of the Lasso regression. The weights depend on estimates of the missingness probabilities (propensity scores) and solve a convex optimization program that trades off bias and variance optimally. Provided that the propensity scores can be pointwise consistently estimated at in-sample data points, our proposed estimator for the regression function is asymptotically normal and semi-parametrically efficient among all asymptotically linear estimators. Furthermore, the proposed estimator keeps its asymptotic properties even if the propensity scores are estimated by modern machine learning techniques. We validate the finite-sample performance of the proposed estimator through comparative simulation studies and the real-world problem of inferring the stellar masses of galaxies in the Sloan Digital Sky Survey., Comment: Substantial revision with some corrected proofs and added experiments. The updated version has 129 pages (32 pages for the main paper), 9 figures, 9 tables

Data centers, the engines of the global Internet, are supported by massive high-speed optical interconnects. In optical fiber communication, the classic direct detection obtains only the intensity of the optical field, while the coherent detection counterpart utilizes both phase and polarization diversities at the expense of beating with a narrow-linewidth and high-stable local oscillator (LO). Herein, we propose and demonstrate a four-dimensional Jones space optical field recovery (4-D JSFR) scheme without LO. The information encoded on the intensity and phase of both polarizations can be captured by the polarization-diversity full-field receiver structure and subsequently extracted through deep neural network-aided field recovery. It achieves similar electrical spectral efficiency as standard intradyne coherent detection. The fully recovered optical field can extend the transmission distance beyond the power fading limitation induced by fiber chromatic dispersion. Furthermore, the LO-free advantage makes 4-D JSFR suitable for monolithic photonic integration, offering a spectrally efficient and cost-effective candidate for large-scale data center applications. Our results could motivate a fundamental paradigm shift in the optical field recovery theory and future optical transceiver design., Comment: 29 pages, 7 figures

The latticework structure known as the cosmic web provides a valuable insight into the assembly history of large-scale structures. Despite the variety of methods to identify the cosmic web structures, they mostly rely on the assumption that galaxies are embedded in a Euclidean geometric space. Here we present a novel cosmic web identifier called SCONCE (Spherical and CONic Cosmic wEb finder) that inherently considers the 2D (RA,DEC) spherical or the 3D (RA,DEC,$z$) conic geometry. The proposed algorithms in SCONCE generalize the well-known subspace constrained mean shift (SCMS) method and primarily address the predominant filament detection problem. They are intrinsic to the spherical/conic geometry and invariant to data rotations. We further test the efficacy of our method with an artificial cross-shaped filament example and apply it to the SDSS galaxy catalogue, revealing that the 2D spherical version of our algorithms is robust even in regions of high declination. Finally, using N-body simulations from Illustris, we show that the 3D conic version of our algorithms is more robust in detecting filaments than the standard SCMS method under the redshift distortions caused by the peculiar velocities of halos. Our cosmic web finder is packaged in python as SCONCE-SCMS and has been made publicly available., Comment: 20 pages, 9 figures, 2 tables

The set of local modes and the ridge lines estimated from a dataset are important summary characteristics of the data-generating distribution. In this work, we consider estimating the local modes and ridges from point cloud data in a product space with two or more Euclidean/directional metric spaces. Specifically, we generalize the well-known (subspace constrained) mean shift algorithm to the product space setting and illuminate some pitfalls in such generalization. We derive the algorithmic convergence of the proposed method, provide practical guidelines on the implementation, and demonstrate its effectiveness on both simulated and real datasets., Comment: 51 pages, 10 figures

This paper studies the linear convergence of the subspace constrained mean shift (SCMS) algorithm, a well-known algorithm for identifying a density ridge defined by a kernel density estimator. By arguing that the SCMS algorithm is a special variant of a subspace constrained gradient ascent (SCGA) algorithm with an adaptive step size, we derive the linear convergence of such SCGA algorithm. While the existing research focuses mainly on density ridges in the Euclidean space, we generalize density ridges and the SCMS algorithm to directional data. In particular, we establish the stability theorem of density ridges with directional data and prove the linear convergence of our proposed directional SCMS algorithm., Comment: Substantial revision. The updated version has 93 pages, 12 figures, and 1 table

The directional mean shift (DMS) algorithm is a nonparametric method for pursuing local modes of densities defined by kernel density estimators on the unit hypersphere. In this paper, we show that any DMS iteration can be viewed as a generalized Expectation-Maximization (EM) algorithm; in particular, when the von Mises kernel is applied, it becomes an exact EM algorithm. Under the (generalized) EM framework, we provide a new proof for the ascending property of density estimates and demonstrate the global convergence of directional mean shift sequences. Finally, we give a new insight into the linear convergence of the DMS algorithm.

Directional data consist of observations distributed on a (hyper)sphere, and appear in many applied fields, such as astronomy, ecology, and environmental science. This paper studies both statistical and computational problems of kernel smoothing for directional data. We generalize the classical mean shift algorithm to directional data, which allows us to identify local modes of the directional kernel density estimator (KDE). The statistical convergence rates of the directional KDE and its derivatives are derived, and the problem of mode estimation is examined. We also prove the ascending property of the directional mean shift algorithm and investigate a general problem of gradient ascent on the unit hypersphere. To demonstrate the applicability of the algorithm, we evaluate it as a mode clustering method on both simulated and real-world data sets., Comment: 92 pages, 11 figures. Accepted to the Journal of Machine Learning Research

Dose reduction in computed tomography (CT) is essential for decreasing radiation risk in clinical applications. Iterative reconstruction is one of the most promising ways to compensate for the increased noise due to reduction of photon flux. Rather than most existing prior-driven algorithms that benefit from manually designed prior functions or supervised learning schemes, in this work we integrate the data-consistency as a conditional term into the iterative generative model for low-dose CT. At the stage of prior learning, the gradient of data density is directly learned from normal-dose CT images as a prior. Then at the iterative reconstruction stage, the stochastic gradient descent is employed to update the trained prior with annealed and conditional schemes. The distance between the reconstructed image and the manifold is minimized along with data fidelity during reconstruction. Experimental comparisons demonstrated the noise reduction and detail preservation abilities of the proposed method.

By acquiring two sets of tomographic measurements at distinct X-ray spectra, the dual-energy CT (DECT) enables quantitative material-specific imaging. However, the conventionally decomposed material basis images may encounter severe image noise amplification and artifacts, resulting in degraded image quality and decreased quantitative accuracy. Iterative DECT image reconstruction algorithms incorporating either the sinogram or the CT image prior information have shown potential advantages in noise and artifact suppression, but with the expense of large computational resource, prolonged reconstruction time, and tedious manual selections of algorithm parameters. To partially overcome these limitations, we develop a domain-transformation enabled end-to-end deep convolutional neural network (DIRECT-Net) to perform high quality DECT material decomposition. Specifically, the proposed DIRECT-Net has immediate accesses to mutual-domain data, and utilizes stacked convolution neural network (CNN) layers for noise reduction and material decomposition. The training data are numerically simulated based on the underlying physics of DECT imaging.The XCAT digital phantom, iodine solutions phantom, and biological specimen are used to validate the performance of DIRECT-Net. The qualitative and quantitative results demonstrate that this newly developed DIRECT-Net is promising in suppressing noise, improving image accuracy, and reducing computation time for future DECT imaging.

Dual-energy computed tomography (DECT) is of great significance for clinical practice due to its huge potential to provide material-specific information. However, DECT scanners are usually more expensive than standard single-energy CT (SECT) scanners and thus are less accessible to undeveloped regions. In this paper, we show that the energy-domain correlation and anatomical consistency between standard DECT images can be harnessed by a deep learning model to provide high-performance DECT imaging from fully-sampled low-energy data together with single-view high-energy data, which can be obtained by using a scout-view high-energy image. We demonstrate the feasibility of the approach with contrast-enhanced DECT scans from 5,753 slices of images of twenty-two patients and show its superior performance on DECT applications. The deep learning-based approach could be useful to further significantly reduce the radiation dose of current premium DECT scanners and has the potential to simplify the hardware of DECT imaging systems and to enable DECT imaging using standard SECT scanners., Comment: 10 pages, 10 figures, 5 tables. Submitted