Interpreting neural networks for biological sequences by learning stochastic masks

Linder, Johannes; Fleur, Alyssa La; Chen, Zibo; Ljubetič, Ajasja; Baker, David C.; Kannan, Sreeram; Seelig, Georg

doi:10.1038/s42256-021-00428-6

Cited by 20 publications

(21 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given the increased performance of a more complex neural network architecture (in this case a deep residual network), we wanted to understand the types of higher-order regulatory features learned by APARENT2 that impact variant effect predictions in polyadenylation signals. To this end, we used a neural network attribution method recently developed by our group—Scrambling—to detect contextual features responsible for the observed variant effects [ 49 ]. To interpret a mutation, we optimize a discretized mask to highlight a shared set of features (nucleotides) in the wildtype and variant sequences that allows reconstruction of their predicted odds ratio when inserted into neutral backgrounds (Fig.…”

Section: Resultsmentioning

confidence: 99%

Deciphering the impact of genetic variation on human polyadenylation using APARENT2

et al. 2022

Self Cite

View full text Add to dashboard Cite

Background 3′-end processing by cleavage and polyadenylation is an important and finely tuned regulatory process during mRNA maturation. Numerous genetic variants are known to cause or contribute to human disorders by disrupting the cis-regulatory code of polyadenylation signals. Yet, due to the complexity of this code, variant interpretation remains challenging. Results We introduce a residual neural network model, APARENT2, that can infer 3′-cleavage and polyadenylation from DNA sequence more accurately than any previous model. This model generalizes to the case of alternative polyadenylation (APA) for a variable number of polyadenylation signals. We demonstrate APARENT2’s performance on several variant datasets, including functional reporter data and human 3′ aQTLs from GTEx. We apply neural network interpretation methods to gain insights into disrupted or protective higher-order features of polyadenylation. We fine-tune APARENT2 on human tissue-resolved transcriptomic data to elucidate tissue-specific variant effects. By combining APARENT2 with models of mRNA stability, we extend aQTL effect size predictions to the entire 3′ untranslated region. Finally, we perform in silico saturation mutagenesis of all human polyadenylation signals and compare the predicted effects of $${>}43$$ > 43 million variants against gnomAD. While loss-of-function variants were generally selected against, we also find specific clinical conditions linked to gain-of-function mutations. For example, we detect an association between gain-of-function mutations in the 3′-end and autism spectrum disorder. To experimentally validate APARENT2’s predictions, we assayed clinically relevant variants in multiple cell lines, including microglia-derived cells. Conclusions A sequence-to-function model based on deep residual learning enables accurate functional interpretation of genetic variants in polyadenylation signals and, when coupled with large human variation databases, elucidates the link between functional 3′-end mutations and human health.

show abstract

Section: Resultsmentioning

confidence: 99%

Deciphering the impact of genetic variation on human polyadenylation using APARENT2

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…Given the increased performance of a more complex network architecture, we wanted to understand the types of higher-order regulatory features learned by APARENT2 that impact variant effect predictions. To this end, we used a neural network attribution method recently developed by our group – Scrambling – to detect contextual features responsible for the observed variant effects [47]. To interpret a mutation, we optimize a discretized attention mask to highlight a shared set of features (nucleotides) in the wildtype-and variant sequences that allows reconstruction of their predicted odds ratio (Figure 2C; see Methods for details).…”

Section: Resultsmentioning

confidence: 99%

“…We adapted our recent work on mask-based interpretation [47] to find the contextual features within a sequence that explain the relative fold change between wildtype- and variant predictions. If the effect of all nucleotides were independent, the solution would simply be to return the mutated position itself (and nothing else).…”

Section: Methodsmentioning

confidence: 99%

“…In-silico interpretation methods have been applied extensively to assess the impact of genetic variants on the underlying cis-regulatory code [42, 37, 30, 43, 44, 45, 46]. Here, we use a mask-based interpretation method for neural networks – Scrambling – to elucidate higher-order features responsible for the predicted variant effects [47]. Specifically, we extend Scramblers to find the minimal set of features which explain the functional differences between a variant and wildtype sequence.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deciphering the Impact of Genetic Variation on Human Polyadenylation

Linder

Kundaje

Seelig

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Genetic variants that disrupt polyadenylation can cause or contribute to genetic disorders. Yet, due to the complex cis-regulation of polyadenylation, variant interpretation remains challenging. Here, we introduce a residual neural network model, APARENT2, that can infer 3’-cleavage and polyadenylation from DNA sequence more accurately than any previous model. This model generalizes to the case of alternative polyadenylation (APA) for a variable number of polyadenylation signals. We demonstrate APARENT2’s performance on several variant datasets, including functional reporter data and human 3’ aQTLs from GTEx. We apply neural network interpretation methods to gain insights into disrupted or protective higher-order features of polyadenylation. We fine-tune APARENT2 on human tissue-resolved transcriptomic data to elucidate tissue-specific variant effects. Finally, we perform in-silico saturation mutagenesis of all human polyadenylation signals and compare the predicted effects of >44 million variants against gnomAD. While loss-of-function variants were generally selected against, we also find specific clinical conditions linked to gain-of-function mutations. For example, using APARENT2’s predictions we detect an association between gain-of-function mutations in the 3’-end and Autism Spectrum Disorder.

show abstract

“…Linder et al improved this technique to use it for various problems such as controlling the level of gene transcription, RNA splicing, or RNA 3' cleavage (Linder et al (2020)). More recently, Linder et al used masks on the sequence to both determine whether each part of the input sequence was sufficient to explain the network predictions and use this information to generate new sequences with similar properties (Linder et al (2021)). Other applications include Cuperus et al who used their trained CNN to predict the translation level of mRNAs from their 5' untranslated sequence (Cuperus et al (2017)).…”

Section: Synthetic Sequence Designmentioning

confidence: 99%

Peer Review #2 of "Genomics enters the deep learning era (v0.1)"

2022

View full text Add to dashboard Cite

Interpreting neural networks for biological sequences by learning stochastic masks

Cited by 20 publications

References 56 publications

Deciphering the impact of genetic variation on human polyadenylation using APARENT2

Deciphering the impact of genetic variation on human polyadenylation using APARENT2

Deciphering the Impact of Genetic Variation on Human Polyadenylation

Peer Review #2 of "Genomics enters the deep learning era (v0.1)"

Contact Info

Product

Resources

About