Toward Identification of Functional Sequences and Variants in Noncoding DNA

Monti, Remo; Ohler, Uwe

doi:10.1146/annurev-biodatasci-122120-110102

Cited by 3 publications

(3 citation statements)

References 136 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1, right). In regulatory genomics, different experiments allow us to determine the functions and characteristics of non-coding DNA regions [22]. This includes, among other methods ChIP-seq, DNase-seq, and ATAC-seq where the enrichment of sequence fragments is measured for various functions like protein binding locations or chromatin accessibility.…”

Section: Introductionmentioning

confidence: 99%

Motif Interactions Affect Post-Hoc Interpretability of Genomic Convolutional Neural Networks

Lemanczyk,

Bartoszewicz,

Renard

2024

Preprint

View full text Add to dashboard Cite

Post-hoc interpretability methods are commonly used to understand decisions of genomic deep learning models and reveal new biological insights. However, interactions between sequence regions (e.g. regulatory elements) impact the learning process as well as interpretability methods that are sensitive to dependencies between features. Since deep learning models learn correlations between the data and output that do not necessarily represent a causal relationship, it is difficult to say how well interacting motif sets are fully captured. Here, we investigate how genomic motif interactions influence model learning and interpretability methods by formalizing possible scenarios where interaction effects appear. This includes the choice of negative data and non-additive effects on the outcome. We generate synthetic data containing interactions for those scenarios and evaluate how they affect the performance of motif detection. We show that post-hoc interpretability methods can miss motifs if interactions are present depending on how negative data is defined. Furthermore, we observe differences in interpretability between additive and non-additive effects as well as between post-hoc interpretability methods.

show abstract

Section: Introductionmentioning

confidence: 99%

Motif Interactions Affect Post-Hoc Interpretability of Genomic Convolutional Neural Networks

Lemanczyk,

Bartoszewicz,

Renard

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…We now know that typical traits are influenced by many loci of small effect; consequently, our inability to fully account for the heritable variance in a trait (“missing heritability”) mainly originates in the statistical limitations of our surveys [22, 6, 36, 18]. Furthermore, there is a high prevalence of significant GWAS hits in the noncoding DNA, especially in regulatory sequences such as enhancers and promoters [46, 23, 26]. The distribution of hits across the genome has been shown to be surprisingly uniform, rather than concentrated in or near “core genes,” i.e., genes with a clear biological and causal link to the trait of interest [12, 38, 49].…”

Section: Introductionmentioning

confidence: 99%

“…We now know that typical traits are influenced by many loci of small effect; consequently, our inability to fully account for the heritable variance in a trait ("missing heritability") mainly originates in the statistical limitations of our surveys [22,6,36,18]. Furthermore, there is a high prevalence of significant GWAS hits in the noncoding DNA, especially in regulatory sequences such as enhancers and promoters [46,23,26].…”

Section: Introductionmentioning

confidence: 99%

Quantitative omnigenic model discovers interpretable genome-wide associations

Ružičková,

Hledík,

Tkačik

2024

Preprint

View full text Add to dashboard Cite

As their statistical power grows, genome-wide association studies (GWAS) have identified an increasing number of loci underlying quantitative traits of interest. These loci are scattered throughout the genome and are individually responsible only for small fractions of the total heritable trait variance. The recently proposed omnigenic model provides a conceptual framework to explain these observations by postulating that numerous distant loci contribute to each complex trait via effect propagation through intracellular regulatory networks. We formalize this conceptual framework by proposing the Quantitative Omnigenic Model (QOM), a statistical model that combines prior knowledge of the regulatory network topology with genomic data. By applying our model to gene expression traits in yeast, we demonstrate that QOM achieves similar gene expression prediction performance to traditional GWAS with hundreds of times less parameters, while simultaneously extracting candidate causal and quantitative chains of effect propagation through the regulatory network for every individual gene. We estimate the fraction of heritable trait variance in cis- and in trans-, break the latter down by effect propagation order, assess the trans- variance not attributable to transcriptional regulation, and show that QOM correctly accounts for the low-dimensional structure of gene expression covariance. We furthermore demonstrate the relevance of QOM for systems biology, by employing it as a statistical test for the quality of regulatory network reconstructions, and linking it to the propagation of non-genetic, environmental effects.

show abstract

Quantitative omnigenic model discovers interpretable genome-wide associations

Ružičková,

Hledík,

Tkačik

2024

Proc. Natl. Acad. Sci. U.S.A.

View full text Add to dashboard Cite

As their statistical power grows, genome-wide association studies (GWAS) have identified an increasing number of loci underlying quantitative traits of interest. These loci are scattered throughout the genome and are individually responsible only for small fractions of the total heritable trait variance. The recently proposed omnigenic model provides a conceptual framework to explain these observations by postulating that numerous distant loci contribute to each complex trait via effect propagation through intracellular regulatory networks. We formalize this conceptual framework by proposing the “quantitative omnigenic model” (QOM), a statistical model that combines prior knowledge of the regulatory network topology with genomic data. By applying our model to gene expression traits in yeast, we demonstrate that QOM achieves similar gene expression prediction performance to traditional GWAS with hundreds of times less parameters, while simultaneously extracting candidate causal and quantitative chains of effect propagation through the regulatory network for every individual gene. We estimate the fraction of heritable trait variance in cis- and in trans- , break the latter down by effect propagation order, assess the trans- variance not attributable to transcriptional regulation, and show that QOM correctly accounts for the low-dimensional structure of gene expression covariance. We furthermore demonstrate the relevance of QOM for systems biology, by employing it as a statistical test for the quality of regulatory network reconstructions, and linking it to the propagation of nontranscriptional (including environmental) effects.

show abstract

Toward Identification of Functional Sequences and Variants in Noncoding DNA

Cited by 3 publications

References 136 publications

Motif Interactions Affect Post-Hoc Interpretability of Genomic Convolutional Neural Networks

Motif Interactions Affect Post-Hoc Interpretability of Genomic Convolutional Neural Networks

Quantitative omnigenic model discovers interpretable genome-wide associations

Quantitative omnigenic model discovers interpretable genome-wide associations

Contact Info

Product

Resources

About