Deciphering the impact of genetic variation on human polyadenylation using APARENT2

• DOI: 10.1186/s13059-022-02799-4
• Publication Date: 2022-11-05T08:02:51Z
• Authors: Johannes Linder; Samantha E. Koplik; Anshul Kundaje; Georg Seelig
• Keywords: 0303 health sciences; QH301-705.5; Autism Spectrum Disorder; Research; RNA Stability; Genetic Variation; Deep learning; QH426-470; Polyadenylation; 03 medical and health sciences; Genetics; RNA; Humans; Biology (General); Untranslated region; Transcriptome; 3' Untranslated Regions; Neural networks; Variant interpretation

Abstract Background 3′-end processing by cleavage and polyadenylation is an important finely tuned regulatory process during mRNA maturation. Numerous genetic variants are known to cause or contribute human disorders disrupting the cis-regulatory code of signals. Yet, due complexity this code, variant interpretation remains challenging. Results We introduce a residual neural network model, APARENT2 , that can infer 3′-cleavage from DNA sequence more accurately than any previous model. This model generalizes case alternative (APA) for variable number demonstrate APARENT2’s performance on several datasets, including functional reporter data 3′ aQTLs GTEx. apply methods gain insights into disrupted protective higher-order features polyadenylation. fine-tune tissue-resolved transcriptomic elucidate tissue-specific effects. By combining with models stability, we extend aQTL effect size predictions entire untranslated region. Finally, perform in silico saturation mutagenesis all signals compare predicted effects $${&gt;}43$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mo>&gt;</mml:mo><mml:mn>43</mml:mn></mml:mrow></mml:math> million against gnomAD. While loss-of-function were generally selected against, also find specific clinical conditions linked gain-of-function mutations. For example, detect association between mutations autism spectrum disorder. To experimentally validate predictions, assayed clinically relevant multiple cell lines, microglia-derived cells. Conclusions A sequence-to-function based deep learning enables accurate and, when coupled large variation databases, elucidates link health.