Flagship Pioneering (United States)

Cryo-electron tomography (cryo-ET) enables the (noisy) observation of individual macromolecular complexes executing their biological functions within a near-native vitrified cellular environment of interest.Subsequent subtomogram averaging (STA) can reveal macromolecular structure at sub-nanometer resolution, though typically at the cost of coarsely discretizing or "averaging out" potentially important heterogeneity.We have previously introduced tomoDRGN, a deep learning tool to resolve ensembles of highly heterogeneous structures imaged via cryo-ET [1].Here we present updates to tomoDRGN aimed at facilitating its use and interaction with modern STA software, and illustrate the resulting insights gained from applying tomoDRGN to experimental cryo-ET datasets.

Spatial transcriptomics is a powerful technology for high-resolution mapping of gene expression in tissue samples, enabling a molecular level understanding of tissue architecture. The acquisition entails dissecting and profiling micron-thick tissue slices, with multiple slices often needed for a comprehensive study. However, the lack of a common coordinate framework (CCF) among slices, due to slicing and displacement variations, can hinder data analysis, making data comparison and integration challenging, and potentially compromising analysis accuracy. Here we present a deep learning algorithm STaCker that unifies the coordinates of transcriptomic slices via an image registration process. STaCker derives a composite image representation by integrating tissue image and gene expression that are transformed to be resilient to noise and batch effects. STaCker overcomes the training data scarcity by training exclusively on diverse synthetic data. Its performance on various benchmarking datasets shows a significant increase in spatial concordance in aligned slices, surpassing existing methods. STaCker also successfully harmonizes multiple real spatial transcriptome datasets acquired from various platforms. These results indicate that STaCker is a valuable computational tool for constructing a CCF with spatial transcriptome data.

Aim: To demonstrate the importance of critical reagent characterization for immunogenicity assay development for multi-specific drugs using two case studies.Methods: Bridging anti-drugantibody (ADA) assay with acid-dissociated samples were used for both cases.Results: In the first case study, the unexpected interference in an ADA assay from clinical samples was identified; a model was created to replicate the issue, and an anti-target antibody was identified to mitigate the target interference. In the second case study, an issue due to non-specific binding of a domain-specific confirmatory reagent was identified, and various mitigation techniques were evaluated.Conclusion: A thorough characterization of the critical reagents helped identify the issues with these ADA case studies and provided strategies for resolving them.

Pre-mRNA splicing, a key process in gene expression, can be therapeutically modulated using various drug modalities, including antisense oligonucleotides (ASOs). However, determining promising targets is impeded by the challenge of systematically mapping splicing-regulatory elements (SREs) in their native sequence context. Here, we use the catalytically dead CRISPR-RfxCas13d RNA-targeting system (dCas13d/gRNA) as a programmable platform to bind SREs and modulate splicing by competing against endogenous splicing factors. SpliceRUSH, a high-throughput screening method, was developed to map SREs in any gene of interest using a lentivirus gRNA library that tiles the genetic region, including distal intronic sequences. When applied to SMN2, a therapeutic target for spinal muscular atrophy, SpliceRUSH robustly identified not only known SREs, but also a novel distal intronic splicing enhancer, which can be targeted to alter exon 7 splicing using either dCas13d/gRNA or ASOs. This technology enables a deeper understanding of splicing regulation with applications for RNA-based drug discovery.

Abstract S-palmitoylation is an essential lipid modification catalysed by zDHHC-palmitoyl acyltransferases that regulates the localisation and activity of substrates in every class of protein and tissue investigated to date. In the heart, S-palmitoylation regulates sodiumcalcium exchanger (NCX1) inactivation, phospholemman (PLM) inhibition of the Na+/K+ ATPase, Nav1.5 influence on membrane excitability and membrane localisation of heterotrimeric G-proteins. The cell surface localised enzyme zDHHC5 palmitoylates NCX1 and PLM and is implicated in injury during anoxia/reperfusion. Information is lacking about how palmitoylation remodels in cardiac diseases. We investigated expression of zDHHC5 in animal models of left ventricular hypertrophy (LVH) and heart failure (HF), along with HF tissue from humans. zDHHC5 expression was rapidly elevated during onset of LVH, whilst HF was associated with decreased zDHHC5 expression. Paradoxically, palmitoylation of the zDHHC5 substrate NCX1 was significantly reduced in LVH but increased in human HF. Overexpression of zDHHC5 in rabbit ventricular cardiomyocytes was not sufficient to drive changes in palmitoylation of zDHHC5 substrates or overall cardiomyocyte contractility, suggesting changes in zDHHC5 expression in disease may not be a primary driver of pathology. zDHHC5 itself is regulated by post-translational modifications, including palmitoylation in its Cterminal tail, and we found the palmitoylation of zDHHC5 may be increased in heart failure in the same manner as NCX1, suggesting additional regulatory mechanisms such as acyl-CoA availability may be involved. Importantly, this study provides the first evidence that palmitoylation of cardiac substrates is altered in the setting of HF, and that expression of zDHHC5 is dysregulated in both hypertrophy and HF.

Particle production from secondary proton-proton collisions, commonly referred to as pile-up, impair the sensitivity of both new physics searches and precision measurements at large hadron collider (LHC) experiments. We propose a novel algorithm, Puma, for modeling pile-up with the help of deep neural networks based on sparse transformers. These attention mechanisms were developed for natural language processing but have become popular in other applications. In a realistic detector simulation, our method outperforms classical benchmark algorithms for pile-up mitigation in key observables. It provides a perspective for mitigating the effects of pile-up in the high luminosity era of the LHC, where up to 200 proton-proton collisions are expected to occur simultaneously.

Relative abundances of bacterial species in the gut microbiome have been linked to many diseases. Species of gut bacteria are ecologically differentiated by their abilities to metabolize different glycans, making glycan delivery a powerful way to alter the microbiome to promote health. Here, we study the properties and therapeutic potential of chemically diverse synthetic glycans (SGs). Fermentation of SGs by gut microbiome cultures results in compound-specific shifts in taxonomic and metabolite profiles not observed with reference glycans, including prebiotics. Model enteric pathogens grow poorly on most SGs, potentially increasing their safety for at-risk populations. SGs increase survival, reduce weight loss, and improve clinical scores in mouse models of colitis. Synthetic glycans are thus a promising modality to improve health through selective changes to the gut microbiome.

Abstract The global clinical artificial intelligence (AI) research landscape is constantly evolving, with heterogeneity across specialties, disease areas, geographical representation, and development maturity. Continual assessment of this landscape is important for monitoring progress. Taking advantage of developments in natural language processing (NLP), we produce an end-to-end NLP pipeline to automate classification and characterization of all original clinical AI research on MEDLINE, outputting real-time results to a public, interactive dashboard (https://aiforhealth.app/).