Biology subjects

Keener, R.

Publications and source records attributed to Keener, R..

8 recordsLinked to original sources

bioRxiv2026

Synthetic MPRA perturbation supervision improves variant-effect prediction in sequence-to-function models

Deep learning sequence-to-function (S2F) models learn to predict functional readouts, such as gene expression, from genetic sequence. However, these models are trained predominantly on the reference genome and struggle to accurately predict the effect of genetic variation on gene expression. Massively parallel reporter assays (MPRAs) explore the direct impact of genetic variation on gene expression, but saturation-mutagenesis MPRAs that measure variant effects directly remain limited, thereby restricting their use for fine-tuning S2F models. In this work, we investigate fine-tuning Borzoi, a state-of-the-art S2F model, on model-generated synthetic MPRA variant-effect data as a scalable alternative to overcome limits on the availability of supervised training data for genetic variation. First, we trained an ensemble of 16 models on 110,359 K562 MPRA measurements. Then, we distilled their predictions into a single student model using MPRA elements associated with 1,000 genes. The distilled model was subsequently used to generate synthetic variant-effect labels across increasing numbers of genes (50 to 1,000), with 600 variants per gene, to fine-tune Borzoi. Performance improved as we included more genes in training. In data-limited settings, training on more genes with fewer variants per gene performed better than training on fewer genes with more variants per gene, given the same number of total variants. Compared to pre-trained Borzoi, Borzoi fine-tuned on 600,000 synthetic samples across 1,000 genes increased Pearson correlation from 0.28 to 0.54 on CAGI5 promoter variants and from 0.39 to 0.46 on 903 GTEx fine-mapped whole-blood expression quantitative trait loci (eQTLs). Importantly, these improvements were achieved while preserving the pre-trained performance across all 7,611 output tracks. Furthermore, fine-tuned Borzoi recovered regulatory motifs identified by the MPRA model in genes unseen during training, suggesting that learned regulatory features can generalize across loci. Together, these results provide a proof of concept that model-generated synthetic supervision can improve variant effect prediction in S2F models while preserving their original pre-trained capabilities, offering a scalable strategy toward capturing the impact of genetic variation on gene regulation.

genomics↗

bioRxiv2025

Fine-tuning sequence to function deep learning models on large-scale proteomic data improves the accuracy of variant effect prediction

Fine-tuning sequence to function models has shown promise for variant effect prediction, but accuracy and generalization to unseen genes and unseen individuals remains a standing challenge. We fine-tuned Borzoi on 54,219 individuals and 2,923 circulating plasma proteins from the UK Biobank Plasma Proteomic Project. Across 150 single-gene models where the genes had a range of cis-heritability we observed that the fine-tuned Borzoi model improved variant effect prediction for 86% of the genes compared to an Elastic Net baseline model. We demonstrated that the improved prediction stems from increased sample size which provides tremendous amounts of rare genetic variants (MAF < 0.01) to the training data. Masking rare and uncommon variants nullified improved performance of fine-tuned Borzoi and we showed that fine-tuned Borzoi highly weights rare variants (MAF < 0.01) while the Elastic Net model highly weights common variants (MAF > 0.05) that are enriched for regulatory regions. We evaluated the generalizability of our model on a fine-tuned Borzoi model trained jointly on varying numbers of genes and observed that these models consistently outperform the pre-trained Borzoi model, the single-gene models yield more accurate results. Together this work demonstrates the importance of including larger sample sizes and rare variants in sequence to function models for variant effect prediction and demonstrates feasibility that these models are capable of highly accurate variant effect prediction.

genetics↗

bioRxiv2024

Human genetic variation reveals FCRL3 is a lymphocyte receptor for Yersinia pestis

Yersinia pestis is the gram-negative bacterium responsible for plague, one of the deadliest and most feared diseases in human history. This bacterium is known to infect phagocytic cells, such as dendritic cells and macrophages, but interactions with non-phagocytic cells of the adaptive immune system are frequently overlooked despite the importance they likely hold for human infection. To discover human genetic determinants of Y. pestis infection, we utilized nearly a thousand genetically diverse lymphoblastoid cell lines in a cellular genome-wide association study method called Hi-HOST (High-throughput Human in-vitrO Susceptibility Testing). We identified a nonsynonymous SNP, rs2282284, in Fc receptor like 3 (FCRL3) associated with bacterial invasion of host cells (p=9x10-8). FCRL3 belongs to the immunoglobulin superfamily and is primarily expressed in lymphocytes. rs2282284 is within a tyrosine-based signaling motif, causing an asparagine-to-serine mutation (N721S) in the most common FCRL3 isoform. Overexpression of FCRL3 facilitated attachment and invasion of non-opsonized Y. pestis. Additionally, FCRL3 colocalized with Y. pestis at sites of cellular attachment, suggesting FCRL3 is a receptor for Y. pestis. These properties were variably conserved across the FCRL family, revealing molecular requirements of attachment and invasion, including an Ig-like C2 domain and a SYK interaction motif. Direct binding was confirmed with purified FCRL5 extracellular domain. Following attachment, invasion of Y. pestis was dependent on SYK and decreased with the N721S mutation. Unexpectedly, this same variant is associated with risk of chronic hepatitis C virus infection in BioBank Japan. Thus, Y. pestis hijacks FCRL proteins, possibly taking advantage of an immune receptor to create a lymphocyte niche during infection.

genetics↗

bioRxiv2024

Sparse matrix factorization of GWAS summary statistics robust to sample sharing improves detection and interpretation of factors with diverse genetic architectures

Complex trait-associated genetic variation is highly pleiotropic. This extensive pleiotropy implies that multi-phenotype analyses are informative for characterizing genetic associations, as they facilitate the discovery of trait-shared and trait-specific variants and pathways ("genetic factors"). Previous efforts have estimated genetic factors using matrix factorization (MF) applied to numerous GWAS. However, existing methods are susceptible to spurious factors arising from residual confounding due to sample-sharing in biobank GWAS. Furthermore, MF approaches have historically estimated dense factors, loaded on most traits and variants, that are challenging to map onto interpretable biological pathways. To address these shortcomings, we introduce "GWAS latent embeddings accounting for noise and regularization" (GLEANR), a MF method for detection of sparse genetic factors from summary statistics. GLEANR accounts for sample sharing between studies and uses regularization to estimate a data-driven number of interpretable factors. GLEANR is robust to confounding induced by shared samples and improves the replication of genetic factors derived from distinct biobanks. We used GLEANR to evaluate 137 diverse GWAS from the UK Biobank, identifying 58 factors that decompose the genetic architecture of input traits and have distinct signatures of negative selection and degrees of polygenicity. These sparse factors can be interpreted with respect to disease, cell-type, and pathway enrichment. We highlight three such factors capturing platelet measure phenotypes and enriched for disease-relevant markers corresponding to distinct stages of platelet differentiation. Overall, GLEANR is a powerful tool for discovering both trait-specific and trait-shared pathways underlying complex traits from GWAS summary statistics.

genomics↗

bioRxiv2024

Genotype inference from aggregated chromatin accessibility data reveals genetic regulatory mechanisms

BackgroundUnderstanding the genetic causes for variability in chromatin accessibility can shed light on the molecular mechanisms through which genetic variants may affect complex traits. Thousands of ATAC-seq samples have been collected that hold information about chromatin accessibility across diverse cell types and contexts, but most of these are not paired with genetic information and come from diverse distinct projects and laboratories. ResultsWe report here joint genotyping, chromatin accessibility peak calling, and discovery of quantitative trait loci which influence chromatin accessibility (caQTLs), demonstrating the capability of performing caQTL analysis on a large scale in a diverse sample set without pre-existing genotype information. Using 10,293 profiling samples representing 1,454 unique donor individuals across 653 studies from public databases, we catalog 23,381 caQTLs in total. After joint discovery analysis, we cluster samples based on accessible chromatin profiles to identify context-specific caQTLs. We find that caQTLs are strongly enriched for annotations of gene regulatory elements across diverse cell types and tissues and are often strongly linked with genetic variation associated with changes in expression (eQTLs), indicating that caQTLs can mediate genetic effects on gene expression. We demonstrate sharing of causal variants for chromatin accessibility and diverse complex human traits, enabling a more complete picture of the genetic mechanisms underlying complex human phenotypes. ConclusionsOur work provides a proof of principle for caQTL calling from previously ungenotyped samples, and represents one of the largest, most diverse caQTL resources currently available, informing mechanisms of genetic regulation of gene expression and contribution to disease.

genomics↗

bioRxiv2024

Aggregation of recount3 RNA-seq data improves the inference of consensus and tissue-specific gene co-expression networks.

BackgroundGene co-expression networks (GCNs) describe relationships among expressed genes key to maintaining cellular identity and homeostasis. However, the small sample size of typical RNA-seq experiments which is several orders of magnitude fewer than the number of genes is too low to infer GCNs reliably. recount3, a publicly available dataset comprised of 316,443 uniformly processed human RNA-seq samples, provides an opportunity to improve power for accurate network reconstruction and obtain biological insight from the resulting networks. ResultsWe compared alternate aggregation strategies to identify an optimal workflow for GCN inference by data aggregation and inferred three consensus networks: a universal network, a non-cancer network, and a cancer network in addition to 27 tissue context-specific networks. Central network genes from our consensus networks were enriched for evolutionarily constrained genes and ubiquitous biological pathways, whereas central context-specific network genes included tissue-specific transcription factors and factorization based on the hubs led to clustering of related tissue contexts. We discovered that annotations corresponding to context-specific networks inferred from aggregated data were enriched for trait heritability beyond known functional genomic annotations and were significantly more enriched when we aggregated over a larger number of samples. ConclusionThis study outlines best practices for network GCN inference and evaluation by data aggregation. We recommend estimating and regressing confounders in each data set before aggregation and prioritizing large sample size studies for GCN reconstruction. Increased statistical power in inferring context-specific networks enabled the derivation of variant annotations that were enriched for concordant trait heritability independent of functional genomic annotations that are context-agnostic. While we observed strictly increasing held-out log-likelihood with data aggregation, we noted diminishing marginal improvements. Future directions aimed at alternate methods for estimating confounders and integrating orthogonal information from modalities such as Hi-C and ChIP-seq can further improve GCN inference.

genomics↗

bioRxiv2023

Validation of human telomere length trans-ancestry meta-analysis association signals identifies POP5 and KBTBD6 as novel human telomere length regulation genes

Telomere length genome-wide association studies (GWAS) have become well-powered to detect novel genes in telomere length regulation. However, no prior work has validated these putative novel genes to confirm the contribution of GWAS loci to telomere length regulation. We conducted a trans-ancestry meta-analysis of 211,369 individuals. Through enrichment analyses of chromatin state and cell-type heritability we identified blood and immune cells as the most relevant cell type to examine telomere length association signals. We validated specific GWAS associations by overexpressing KBTBD6, a component of an E3 ubiquitin ligase complex, and POP5, a component of the Ribonuclease P/MRP complex, and demonstrating that both lengthened telomeres as predicted by our statistical analyses. CRISPR/Cas9 deletion of the predicted causal regions of these association peaks in K562 immortalized blood cells reduced expression of these genes, demonstrating that these loci are related to transcriptional regulation of KBTBD6 and POP5, respectively. Together our results demonstrate the utility of telomere length GWAS in the identification of novel telomere length regulation mechanisms and highlight the importance of the proteasome-ubiquitin pathway in telomere length regulation.

genetics↗

bioRxiv2019

Tel1 activation by the MRX complex is sufficient for telomere length regulation but not for the DNA damage response in S. cerevisiae

Previous models suggested that regulation of telomere length in S. cerevisiae by Tel1(ATM) and Mec1(ATR) parallel the established pathways regulating the DNA damage response. Here we provide evidence that telomere length regulation differs from the DNA damage response in both the Tel1 and Mec1 pathways. We found that Rad53 mediates a Mec1 telomere length regulation pathway but is dispensable for Tel1 telomere length regulation, whereas in the DNA damage response Rad53 is regulated by both Mec1 and Tel1. Using epistasis analysis with a Tel1 hypermorphic allele, Tel1-hy909, we found that the MRX complex is not required downstream of Tel1 for telomere elongation but is required downstream of Tel1 for the DNA damage response. Since models that invoke a required end processing event for telomerase elongation are primarily based on the yeast pathways, our data call for a re-examination of the requirement for telomere end processing in both yeast and mammalian cells.

molecular biology↗