Publications

The borylation of aryl and heteroaryl C-H bonds is valuable for the site-selective functionalization of C-H bonds in complex molecules. Iridium catalysts ligated by bipyridine ligands catalyze the borylation of the C-H bond that is most acidic and least sterically hindered in an arene, but predicting the site of borylation in molecules containing multiple arenes is difficult. To address this challenge, we report a hybrid computational model that predicts the Site of Borylation (SoBo) in complex molecules. The SoBo model combines density functional theory, semiempirical quantum mechanics, cheminformatics, linear regression, and machine learning to predict site selectivity and to extrapolate these predictions to new chemical space. Experimental validation of SoBo showed that the model predicts the major site of borylation of pharmaceutical intermediates with higher accuracy than prior machine-learning models or human experts, demonstrating that SoBo will be useful to guide experiments for the borylation of specific C(sp²)-H bonds during pharmaceutical development.

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.

ABSTRACT A novel polyphosphate kinase (PPK) was retrieved from an uncultivated organism in activated sludge carrying out enhanced biological phosphorus removal (EBPR). Acetate-fed laboratory-scale sequencing batch reactors were used to maintain sludge with a high phosphorus content (approximately 11% of the biomass). PCR-based clone libraries of small subunit rRNA genes and fluorescent in situ hybridization (FISH) were used to verify that the sludge was enriched in Rhodocyclus -like β- Proteobacteria known to be associated with sludges carrying out EBPR. These organisms comprised approximately 80% of total bacteria in the sludge, as assessed by FISH. Degenerate PCR primers were designed to retrieve fragments of putative ppk genes from a pure culture of Rhodocyclus tenuis and from organisms in the sludge. Four novel ppk homologs were found in the sludge, and two of these (types I and II) shared a high degree of amino acid similarity with R. tenuis PPK (86 and 87% similarity, respectively). Dot blot analysis of total RNA extracted from sludge demonstrated that the Type I ppk mRNA was present, indicating that this gene is expressed during EBPR. Inverse PCR was used to obtain the full Type I sequence from sludge DNA, and a full-length PPK was cloned, overexpressed, and purified to near homogeneity. The purified PPK has a specific activity comparable to that of other PPKs, has a requirement for Mg 2+ , and does not appear to operate in reverse. PPK activity was found mainly in the particulate fraction of lysed sludge microorganisms.

As the demand for high-quality data in model training grows, researchers and developers are increasingly generating synthetic data to tune and train LLMs. However, current data generation methods rely on seed sets containing tens of thousands of examples to prompt instruction-tuned models. This reliance can be especially problematic when the curation of high-quality examples is expensive or difficult. In this paper we explore the novel few-shot synthetic data generation setting -- generating a high-quality dataset from a few examples. We show that when working with only a few seed examples, instruction-tuned models used in current synthetic data methods produce insufficient diversity for downstream tasks. In contrast, we show that base models without post-training, largely untapped for synthetic data generation, offer substantially greater output diversity, albeit with lower instruction following abilities. Leveraging this insight, we propose Base-Refine (BARE), a novel two-stage method that combines the diversity of base models with the quality assurance of instruction-tuned models. BARE excels in few-shot synthetic data generation: using only 3 seed examples it generates diverse, high-quality datasets that significantly improve downstream task performance. We show that fine-tuning Llama 3.1 8B with 1,000 BARE-generated samples achieves performance comparable to state-of-the-art similarly sized models on LiveCodeBench tasks. Furthermore, data generated with BARE enables a 101% improvement for a fine-tuned Llama 3.2 1B on GSM8K over data generated by only instruction-models, and an 18.4% improvement for a fine-tuned Llama 3.1 8B over the state-of-the-art RAFT method for RAG data generation.

We developed a new electrochemically active Li storage material based on a polyanionic framework. By replacing a portion of Mn with Fe and Mg in monoclinic lithium manganese borate (LiMnBO 3 ), a phase-pure LiMn 0.5 Fe 0.4 Mg 0.1 BO 3 was successfully synthesized. A C/50-rate galvanostatic discharge delivered a near theoretical capacity of 201 mAh g -1 at room temperature with much improved rate capability, 120 mAh g -1 at a 1C discharge, compared to other borate chemistries. We explained this superior Li storage activity in relation to the topological effect of substitution, which leads to facile Li diffusion and stabilizes the structure. Combining with the inherent stability of the borate group against oxygen evolution, we believe that the enhanced electrochemical properties achieved in this study makes the LiMn 0.5 Fe 0.4 Mg 0.1 BO 3 compound a strong contender as a new cathode material for Li-ion batteries and will bring more attention to the lithium transition metal borate chemistry.

While there have been significant gains in the field of automated video description, the generalization performance of automated description models to novel domains remains a major barrier to using these systems in the real world. Most visual description methods are known to capture and exploit patterns in the training data leading to evaluation metric increases, but what are those patterns? In this work, we examine several popular visual description datasets, and capture, analyze, and understand the dataset-specific linguistic patterns that models exploit but do not generalize to new domains. At the token level, sample level, and dataset level, we find that caption diversity is a major driving factor behind the generation of generic and uninformative captions. We further show that state-of-the-art models even outperform held-out ground truth captions on modern metrics, and that this effect is an artifact of linguistic diversity in datasets. Understanding this linguistic diversity is key to building strong captioning models, we recommend several methods and approaches for maintaining diversity in the collection of new data, and dealing with the consequences of limited diversity when using current models and metrics.

Abstract Achieving high titer rates and yields (TRY) remains a bottleneck in the production of heterologous products through microbial systems, requiring elaborate engineering and many iterations. Reliable scaling of engineered strains is also rarely addressed in the first designs of the engineered strains. Both high TRY and scale are challenging metrics to achieve due to the inherent trade-off between cellular use of carbon towards growth vs. target metabolite production. We hypothesized that being able to strongly couple product formation with growth may lead to improvements across both metrics. In this study, we use elementary mode analysis to predict metabolic reactions that could be targeted to couple the production of indigoidine, a sustainable pigment, with the growth of the chosen host, Pseudomonas putida KT2440. We then filtered the set of 16 predicted reactions using -omics data. We implemented a total of 14 gene knockdowns using a CRISPRi method optimized for P. putida and show that the resulting engineered P. putida strain could achieve high TRY. The engineered pairing of product formation with carbon use also shifted production from stationary to exponential phase and the high TRY phenotype was maintained across scale. In one design cycle, we constructed an engineered P. putida strain that demonstrates close to 50% maximum theoretical yield (0.33 g indigoidine/g glucose consumed), reaching 25.6 g/L indigoidine and a rate of 0.22g/l/h in exponential phase. These desirable phenotypes were maintained from batch to fed-batch cultivation mode, and from 100ml shake flasks to 250 mL ambr® and 2 L bioreactors.

Medium- and high-entropy alloys based on the CrCoNi-system have been shown to display outstanding strength, tensile ductility and fracture toughness (damage-tolerance properties), especially at cryogenic temperatures. Here we examine the JIc and (back-calculated) KJIc fracture toughness values of the face-centered cubic, equiatomic CrCoNi and CrMnFeCoNi alloys at 20 K. At flow stress values of ~1.5 GPa, crack-initiation KJIc toughnesses were found to be exceptionally high, respectively 235 and 415 MPa(square-root)m for CrMnFeCoNi and CrCoNi, with the latter displaying a crack-growth toughness Kss exceeding 540 MPa(square-root)m after 2.25 mm of stable cracking, which to our knowledge is the highest such value ever reported. Characterization of the crack-tip regions in CrCoNi by scanning electron and transmission electron microscopy reveal deformation structures at 20 K that are quite distinct from those at higher temperatures and involve heterogeneous nucleation, but restricted growth, of stacking faults and fine nano-twins, together with transformation to the hexagonal closed-packed phase. The coherent interfaces of these features can promote both the arrest and transmission of dislocations to generate respectively strength and ductility which strongly contributes to sustained strain hardening. Indeed, we believe that these nominally single-phase, concentrated solid-solution alloys develop their fracture resistance through a progressive synergy of deformation mechanisms, including dislocation glide, stacking-fault formation, nano-twinning and eventually in situ phase transformation, all of which serve to extend continuous strain hardening which simultaneously elevates strength and ductility (by delaying plastic instability), leading to truly exceptional resistance to fracture.