Networks in machine learning offer examples of complex high-dimensional dynamical systems inspired by and reminiscent of biological systems. Here, we study the learning dynamics of generalized Hopfield networks, which permit visualization of internal memories. These networks have been shown to proceed through a “feature-to-prototype” transition, as the strength of network nonlinearity is increased, wherein the learned, or terminal, states of internal memories transition from mixed to pure states. Focusing on the prototype learning dynamics of the internal memories, we observe stereotypical dynamics of memories wherein similar subgroups of memories sequentially split at well-defined saddles. The splitting order is interpretable and reproducible from one simulation to the other. The dynamics prior to splits are robust to variations in many features of the system. To develop a more rigorous understanding of these global dynamics, we study smaller subsystems that exhibit similar properties to the full system. Within these smaller systems, we combine analytical calculations with numerical simulations to study the dynamics of the feature-to-prototype transition, and the emergence of saddle points in the learning landscape. We exhibit regimes where saddles appear and disappear through saddle-node bifurcations, qualitatively changing the distribution of learned memories as the strength of the nonlinearity is varied—allowing us to systematically investigate the mechanisms that underlie the emergence of the learning dynamics. Several features of the learning dynamics are reminiscent of the Waddington's caricature of cellular differentiation, and we attempt to make this analogy more precise. Memories can thus differentiate in a predictive and controlled way, revealing bridges between experimental biology, dynamical systems theory, and machine learning. Published by the American Physical Society 2024

Transcriptional regulation, involving the complex interplay between regulatory sequences and proteins, directs all biological processes. Computational models of transcription lack generalizability to accurately extrapolate in unseen cell types and conditions. Here, we introduce GET, an interpretable foundation model designed to uncover regulatory grammars across 213 human fetal and adult cell types. Relying exclusively on chromatin accessibility data and sequence information, GET achieves experimental-level accuracy in predicting gene expression even in previously unseen cell types. GET showcases remarkable adaptability across new sequencing platforms and assays, enabling regulatory inference across a broad range of cell types and conditions, and uncovering universal and cell type specific transcription factor interaction networks. We evaluated its performance on prediction of regulatory activity, inference of regulatory elements and regulators, and identification of physical interactions between transcription factors. Specifically, we show GET outperforms current models in predicting lentivirus-based massive parallel reporter assay readout with reduced input data. In fetal erythroblasts, we identify distal (>1Mbp) regulatory regions that were missed by previous models. In B cells, we identified a lymphocyte-specific transcription factor-transcription factor interaction that explains the functional significance of a leukemia-risk predisposing germline mutation. In sum, we provide a generalizable and accurate model for transcription together with catalogs of gene regulation and transcription factor interactions, all with cell type specificity.

We introduce mvSuSiE, a multi-trait fine-mapping method for identifying putative causal variants from genetic association data (individual-level or summary data). mvSuSiE learns patterns of shared genetic effects from data, and exploits these patterns to improve power to identify causal SNPs. Comparisons on simulated data show that mvSuSiE is competitive in speed, power and precision with existing multi-trait methods, and uniformly improves on single-trait fine-mapping (SuSiE) in each trait separately. We applied mvSuSiE to jointly fine-map 16 blood cell traits using data from the UK Biobank. By jointly analyzing the traits and modeling heterogeneous effect sharing patterns, we discovered a much larger number of causal SNPs (>3,000) compared with single-trait fine-mapping, and with narrower credible sets. mvSuSiE also more comprehensively characterized the ways in which the genetic variants affect one or more blood cell traits; 68% of causal SNPs showed significant effects in more than one blood cell type.

Apart from ancestry, personal or environmental covariates may contribute to differences in polygenic score (PGS) performance. We analyzed effects of covariate stratification and interaction on body mass index (BMI) PGS (PGS BMI ) across four cohorts of European (N=491,111) and African (N=21,612) ancestry. Stratifying on binary covariates and quintiles for continuous covariates, 18/62 covariates had significant and replicable R 2 differences among strata. Covariates with the largest differences included age, sex, blood lipids, physical activity, and alcohol consumption, with R 2 being nearly double between best and worst performing quintiles for certain covariates. 28 covariates had significant PGS BMI -covariate interaction effects, modifying PGS BMI effects by nearly 20% per standard deviation change. We observed overlap between covariates that had significant R 2 differences among strata and interaction effects – across all covariates, their main effects on BMI were correlated with their maximum R 2 differences and interaction effects (0.56 and 0.58, respectively), suggesting high-PGS BMI individuals have highest R 2 and increase in PGS effect. Using quantile regression, we show the effect of PGS BMI increases as BMI itself increases, and that these differences in effects are directly related to differences in R 2 when stratifying by different covariates. Given significant and replicable evidence for context-specific PGS BMI performance and effects, we investigated ways to increase model performance taking into account non-linear effects. Machine learning models (neural networks) increased relative model R 2 (mean 23%) across datasets. Finally, creating PGS BMI directly from GxAge GWAS effects increased relative R 2 by 7.8%. These results demonstrate that certain covariates, especially those most associated with BMI, significantly affect both PGS BMI performance and effects across diverse cohorts and ancestries, and we provide avenues to improve model performance that consider these effects.

Subcutaneous (s.c.) administration of monoclonal antibodies (mAbs) can reduce treatment burden for patients and healthcare systems compared with intravenous (i.v.) infusion through shorter administration times, made possible by convenient, patient-centric devices. A deeper understanding of clinical pharmacology principles related to efficacy and safety of s.c.-administered mAbs over the past decade has streamlined s.c. product development. This review presents learnings from key constituents of the s.c. mAb development pathway, including pharmacology, administration variables, immunogenicity, and delivery devices. Restricted mAb transportation through the hypodermis explains their incomplete absorption at a relatively slow rate (pharmacokinetic (PK)) and may impact mAb-cellular interactions and/or onset and magnitude of physiological responses (pharmacodynamic). Injection volumes, formulation, rate and site of injection, and needle attributes may affect PKs and the occurrence/severity of adverse events like injection-site reactions or pain, with important consequences for treatment adherence. A review of immunogenicity data for numerous compounds reveals that incidence of anti-drug antibodies (ADAs) is generally comparable across i.v. and s.c. routes, and complementary factors including response magnitude (ADA titer), persistence over time, and neutralizing antibody presence are needed to assess clinical impact. Finally, four case studies showcase how s.c. biologics have been clinically developed: (i) by implementation of i.v./s.c. bridging strategies to streamline PD-1/PD-L1 inhibitor development, (ii) through co-development with i.v. presentations for anti-severe acute respiratory syndrome-coronavirus 2 antibodies to support rapid deployment of both formulations, (iii) as the lead route for bispecific T cell engagers (BTCEs) to mitigate BTCE-mediated cytokine release syndrome, and (iv) for pediatric patients in the case of dupilumab.

AbstractLow-coverage imputation is becoming ever more present in ancient DNA (aDNA) studies. Imputation pipelines commonly used for present-day genomes have been shown to yield accurate results when applied to ancient genomes. However,post-mortemdamage (PMD), in the form of C-to-T substitutions at the reads termini, and contamination with DNA from closely related species can potentially affect imputation performance in aDNA. In this study, we evaluated imputation performance i) when using a genotype caller designed for aDNA, ATLAS, compared to bcftools, and ii) when contamination is present. We evaluated imputation performance with principal component analyses (PCA) and by calculating imputation error rates. With a particular focus on differently imputed sites, we found that using ATLAS prior to imputation substantially improved imputed genotypes for a very damaged ancient genome (42% PMD). For the remaining genomes, ATLAS brought limited gains. Finally, to examine the effect of contamination on imputation, we added various amounts of reads from two present-day genomes to a previously downsampled high-coverage ancient genome. We observed that imputation accuracy drastically decreased for contamination rates above 5%. In conclusion, we recommend i) accounting for PMD by using a genotype caller such as ATLAS before imputing highly damaged genomes and ii) only imputing genomes containing up to 5% of contamination.

Immunoglobulin E (IgE) is a key driver of type 1 hypersensitivity reactions and allergic disorders, which are globally increasing in number and severity. Although eliminating pathogenic IgE may be a powerful way to treat allergy, no therapeutic strategy reported to date can fully ablate IgE production. Interleukin-4 receptor α (IL-4Rα) signaling is required for IgE class switching, and IL-4Rα blockade gradually reduces, but does not eliminate, IgE. The persistence of IgE after IL-4Rα blockade may be due to long-lived IgE+ plasma cells that maintain serological memory to allergens and thus may be susceptible to plasma cell-targeted therapeutics. We demonstrate that transient administration of a B cell maturation antigen x CD3 (BCMAxCD3) bispecific antibody markedly depletes IgE, as well as other immunoglobulins, by ablating long-lived plasma cells, although IgE and other immunoglobulins rapidly rebound after treatment. Concomitant IL-4Rα blockade specifically and durably prevents the reemergence of IgE by blocking IgE class switching while allowing the restoration of other immunoglobulins. Moreover, this combination treatment prevented anaphylaxis in mice. Together with additional cynomolgus monkey and human data, our studies demonstrate that allergic memory is primarily maintained by both non-IgE+ memory B cells that require class switching and long-lived IgE+ plasma cells. Our combination approach to durably eliminate pathogenic IgE has potential to benefit allergy in humans while preserving antibody-mediated immunity.

GDF15, a hormone acting on the brainstem, has been implicated in the nausea and vomiting of pregnancy, including its most severe form, hyperemesis gravidarum (HG), but a full mechanistic understanding is lacking1-4. Here we report that fetal production of GDF15 and maternal sensitivity to it both contribute substantially to the risk of HG. We confirmed that higher GDF15 levels in maternal blood are associated with vomiting in pregnancy and HG. Using mass spectrometry to detect a naturally labelled GDF15 variant, we demonstrate that the vast majority of GDF15 in the maternal plasma is derived from the feto-placental unit. By studying carriers of rare and common genetic variants, we found that low levels of GDF15 in the non-pregnant state increase the risk of developing HG. Conversely, women with β-thalassaemia, a condition in which GDF15 levels are chronically high5, report very low levels of nausea and vomiting of pregnancy. In mice, the acute food intake response to a bolus of GDF15 is influenced bi-directionally by prior levels of circulating GDF15 in a manner suggesting that this system is susceptible to desensitization. Our findings support a putative causal role for fetally derived GDF15 in the nausea and vomiting of human pregnancy, with maternal sensitivity, at least partly determined by prepregnancy exposure to the hormone, being a major influence on its severity. They also suggest mechanism-based approaches to the treatment and prevention of HG.