Publications

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

Shark is a new data analysis system that marries query processing with complex analytics on large clusters.It leverages a novel distributed memory abstraction to provide a unified engine that can run SQL queries and sophisticated analytics functions (e.g., iterative machine learning) at scale, and efficiently recovers from failures mid-query.This allows Shark to run SQL queries up to 100 faster than Apache Hive, and machine learning programs up to 100 faster than Hadoop.Unlike previous systems, Shark shows that it is possible to achieve these speedups while retaining a MapReduce-like execution engine, and the fine-grained fault tolerance properties that such engines provide.It extends such an engine in several ways, including column-oriented in-memory storage and dynamic mid-query replanning, to effectively execute SQL.The result is a system that matches the speedups reported for MPP analytic databases over MapReduce, while offering fault tolerance properties and complex analytics capabilities that they lack.

Abstract Machine learning (ML) has revolutionized drug discovery by enabling the exploration of vast, uncharted chemical spaces essential for discovering novel patentable drugs. Despite the critical role of human G protein-coupled receptors (hGPCRs) in FDA-approved drugs, exhaustive in-distribution drug-target interaction (DTI) testing across all pairs of hGPCRs and known drugs is rare due to significant economic and technical challenges. This often leaves off-target effects unexplored, which poses a considerable risk to drug safety. In contrast to the traditional focus on out-of-distribution (OOD) exploration (drug discovery), we introduce a neighborhood-to-prediction model termed Chemical Space Neural Networks (CSNN) that leverages network homophily and training-free graph neural networks (GNNs) with Labels as Features (LaF). We show that CSNN’s ability to make accurate predictions strongly correlates with network homophily. Thus, LaFs strongly increase a ML model’s capacity to enhance in-distribution prediction accuracy, which we show by integrating labeled data during inference. We validate these advancements in a high-throughput yeast biosensing system (3773 DTIs, 539 compounds, 7 hGPCRs) to discover novel DTIs for FDA-approved drugs and to expand the general understanding of how to build reliable predictors to guide experimental verification.

The tooth enamel of vertebrates comprises a hyper-mineralized bioceramic, but is distinguished by an exceptional durability to resist impact and wear throughout the lifetime of organisms; however, enamels exhibit a low resistance to the initiation of large-scale cracks comparable to that of geological minerals based on fracture mechanics. Here we reveal that the tooth enamel, specifically from the giant panda, is capable of developing durability through counteracting the early stage of damage by partially recovering its innate geometry and structure at nano- to micro- length-scales autonomously. Such an attribute results essentially from the unique architecture of tooth enamel, specifically the vertical alignment of nano-scale mineral fibers and micro-scale prisms within a water-responsive organic-rich matrix, and can lead to a decrease in the dimension of indent damage in enamel introduced by indentation. Hydration plays an effective role in promoting the recovery process and improving the indentation fracture toughness of enamel (by ∼73%), at a minor cost of micro-hardness (by ∼5%), as compared to the dehydrated state. The nano-scale mechanisms that are responsible for the recovery deformation, specifically the reorientation and rearrangement of mineral fragments and the inter- and intra-prismatic sliding between constituents that are closely related to the viscoelasticity of organic matrix, are examined and analyzed with respect to the structure of tooth enamel. Our study sheds new light on the strategies underlying Nature’s design of durable ceramics which could be translated into man-made systems in developing high-performance ceramic materials. Statement of Significance Tooth enamel plays a critical role in the function of teeth by providing a hard surface layer to resist wear/impact throughout the lifetime of organisms; however, such enamel exhibits a remarkably low resistance to the initiation of large-scale cracks, of hundreds of micrometers or more, comparable to that of geological minerals. Here we reveal that tooth enamel, specifically that of the giant panda, is capable of partially recovering its geometry and structure to counteract the early stages of damage at nano- to micro-scale dimensions autonomously. Such an attribute results essentially from the architecture of enamel but is markedly enhanced by hydration. Our work discerns a series of mechanisms that lead to the deformation and recovery of enamel and identifies a unique source of durability in the enamel to accomplish this function. The ingenious design of tooth enamel may inspire the development of new durable ceramic materials in man-made systems.

CO2 valorization is aimed at converting waste CO2 to value-added products. While steady progress has been achieved through diverse catalytic strategies, including CO2 electrosynthesis, CO2 thermocatalysis, and biological CO2 fixation, each of these approaches have distinct limitations. Inorganic catalysts only enable synthesis beyond C2 and C3 products with poor selectivity and with a high energy requirement. Meanwhile, although biological organisms can selectively produce complex products from CO2, their slow autotrophic metabolism limits their industrial feasibility. Here, we present an abiotic approach leveraging electrochemical and thermochemical catalysis to complete the conversion of CO2 to life-sustaining carbohydrate sugars akin to photosynthesis. CO2 was electrochemically converted to glycolaldehyde and formaldehyde using copper nanoparticles and boron-doped diamond cathodes, respectively. CO2-derived glycolaldehyde then served as the key autocatalyst for the formose reaction, where glycolaldehyde and formaldehyde combined in the presence of an alkaline earth metal catalyst to form a variety of C4 - C8 sugars, including glucose. In turn, these sugars were used as a feedstock for fast-growing and genetically modifiable Escherichia coli. Altogether, we have assembled a platform that pushes the boundaries of product complexity achievable from CO2 conversion while demonstrating CO2 integration into life-sustaining sugars.