Tony Zeng

Noncoding regulatory elements in the human genome can control gene expression through distinct mechanisms, including acting as enhancers, silencers, or topological elements. However, we lack tools to identify all such classes systematically and quantify their effects on gene expression in an endogenous genomic context. Here we develop Swap-seq, a method that combines pooled twin prime editing with flow-sorting to delete hundreds of endogenous regulatory sequences and measure their quantitative effects on the expression of a nearby gene. We apply Swap-seq to dissect the regulatory elements that control PPIF expression in THP-1 monocytes. We quantify the effects of distal and intronic enhancers, characterize CTCF insulators, and discover a silencer that regulates distal genes through histone deacetylase (HDAC)-mediated repression of neighboring enhancers. Hundreds of other genomic elements share its chromatin signatures and transcription factor binding sites, suggesting that this silencer represents a broader class of repressive elements. We leverage Swap-seq data to benchmark sequence-to-function models and show that they often capture the local activity of regulatory elements but fail to link them to distal target genes. Swap-seq thus provides a generalizable tool to discover regulatory elements across mechanistic classes, characterize their effects on gene expression, and generate large-scale datasets for evaluating sequence-to-function models.

Genetic association studies provide a unique tool for identifying candidate causal links from genes to human traits and diseases. However, it is challenging to determine the biological mechanisms underlying most associations, and we lack genome-scale approaches for inferring causal mechanistic pathways from genes to cellular functions to traits. Here we propose approaches to bridge this gap by combining quantitative estimates of gene–trait relationships from loss-of-function burden tests¹ with gene-regulatory connections inferred from Perturb-seq experiments² in relevant cell types. By combining these two forms of data, we aim to build causal graphs in which the directional associations of genes with a trait can be explained by their regulatory effects on biological programs or direct effects on the trait³. As a proof of concept, we constructed a causal graph of the gene-regulatory hierarchy that jointly controls three partially co-regulated blood traits. We propose that perturbation studies in trait-relevant cell types, coupled with gene-level effect sizes for traits, can bridge the gap between genetic association and biological mechanism.

: Genetic association studies provide a unique tool for identifying causal links from genes to human traits and diseases. However, it is challenging to determine the biological mechanisms underlying most associations, and we lack genome-scale approaches for inferring causal mechanistic pathways from genes to cellular functions to traits. Here we propose new approaches to bridge this gap by combining quantitative estimates of gene-trait relationships from loss-of-function burden tests with gene-regulatory connections inferred from Perturb-seq experiments in relevant cell types. By combining these two forms of data, we aim to build causal graphs in which the directional associations of genes with a trait can be explained by their regulatory effects on biological programs or direct effects on the trait. As a proof-of-concept, we constructed a causal graph of the gene regulatory hierarchy that jointly controls three partially co-regulated blood traits. We propose that perturbation studies in trait-relevant cell types, coupled with gene-level effect sizes for traits, can bridge the gap between genetics and biology.

Genome-wide association studies (GWASs) have identified tens of thousands of disease associated variants and provided critical insights into developing effective treatments. However, limited sample sizes have hindered the discovery of variants for uncommon and rare diseases. Here, we introduce KGWAS, a novel geometric deep learning method that leverages a massive functional knowledge graph across variants and genes to improve detection power in small-cohort GWASs significantly. KGWAS assesses the strength of a variant’s association to disease based on the aggregate GWAS evidence across molecular elements interacting with the variant within the knowledge graph. Comprehensive simulations and replication experiments showed that, for small sample sizes ( N =1-10K), KGWAS identified up to 100% more statistically significant associations than state-of-the-art GWAS methods and achieved the same statistical power with up to 2.67× fewer samples. We applied KGWAS to 554 uncommon UK Biobank diseases ( N _case case LRRC32 expression in CD4+ regulatory T cells, and rs7312765 (on 12q12) associated with the rare disease myasthenia gravis potentially via regulating PPHLN1 expression in neuron-related cell types. Furthermore, KGWAS consistently improves downstream analyses such as identifying disease-specific network links for interpreting GWAS variants, identifying disease-associated genes, and identifying disease-relevant cell populations. Overall, KGWAS is a flexible and powerful AI model that integrates growing functional genomics data to discover novel variants, genes, cells, and networks, especially valuable for small cohort diseases.

Measures of selective constraint on genes have been used for many applications, including clinical interpretation of rare coding variants, disease gene discovery and studies of genome evolution. However, widely used metrics are severely underpowered at detecting constraints for the shortest ~25% of genes, potentially causing important pathogenic mutations to be overlooked. Here we developed a framework combining a population genetics model with machine learning on gene features to enable accurate inference of an interpretable constraint metric, s_het. Our estimates outperform existing metrics for prioritizing genes important for cell essentiality, human disease and other phenotypes, especially for short genes. Our estimates of selective constraint should have wide utility for characterizing genes relevant to human disease. Finally, our inference framework, GeneBayes, provides a flexible platform that can improve the estimation of many gene-level properties, such as rare variant burden or gene expression differences.

Measures of selective constraint on genes have been used for many applications including clinical interpretation of rare coding variants, disease gene discovery, and studies of genome evolution. However, widely-used metrics are severely underpowered at detecting constraint for the shortest ∼25% of genes, potentially causing important pathogenic mutations to be overlooked. We developed a framework combining a population genetics model with machine learning on gene features to enable accurate inference of an interpretable constraint metric, s_het. Our estimates outperform existing metrics for prioritizing genes important for cell essentiality, human disease, and other phenotypes, especially for short genes. Our new estimates of selective constraint should have wide utility for characterizing genes relevant to human disease. Finally, our inference framework, GeneBayes, provides a flexible platform that can improve estimation of many gene-level properties, such as rare variant burden or gene expression differences.

The Discrete-Time Wright Fisher (DTWF) model and its large population diffusion limit are central to population genetics. These models describe the forward-in-time evolution of the frequency of an allele in a population and can include the fundamental forces of genetic drift, mutation, and selection. Computing like-lihoods under the diffusion process is feasible, but the diffusion approximation breaks down for large sample sizes or in the presence of strong selection. Unfortunately, existing methods for computing likelihoods under the DTWF model do not scale to current exome sequencing sample sizes in the hundreds of thousands. Here we present an algorithm that approximates the DTWF model with provably bounded error and runs in time linear in the size of the population. Our approach relies on two key observations about Binomial distributions. The first is that Binomial distributions are approximately sparse. The second is that Binomial distributions with similar success probabilities are extremely close as distributions, allowing us to approximate the DTWF Markov transition matrix as a very low rank matrix. Together, these observations enable matrix-vector multiplication in linear (as opposed to the usual quadratic) time. We prove similar properties for Hypergeometric distributions, enabling fast computation of likelihoods for subsamples of the population. We show theoretically and in practice that this approximation is highly accurate and can scale to population sizes in the billions, paving the way for rigorous biobank-scale population genetic inference. Finally, we use our results to estimate how increasing sample sizes will improve the estimation of selection coefficients acting on loss-of-function variants. We find that increasing sample sizes beyond existing large exome sequencing cohorts will provide essentially no additional information except for genes with the most extreme fitness effects.

Measures of selective constraint on genes have been used for many applications including clinical interpretation of rare coding variants, disease gene discovery, and studies of genome evolution. However, widely-used metrics are severely underpowered at detecting constraint for the shortest ~25% of genes, potentially causing important pathogenic mutations to be over-looked. We developed a framework combining a population genetics model with machine learning on gene features to enable accurate inference of an interpretable constraint metric, s _het . Our estimates outperform existing metrics for prioritizing genes important for cell essentiality, human disease, and other phenotypes, especially for short genes. Our new estimates of selective constraint should have wide utility for characterizing genes relevant to human disease. Finally, our inference framework, GeneBayes, provides a flexible platform that can improve estimation of many gene-level properties, such as rare variant burden or gene expression differences.