PAL-AI reveals genetic determinants that control poly(A)-tail length during oocyte maturation, with relevance to human infertility

In oocytes of mammals and other animals, gene regulation is mediated primarily through changes in poly(A)-tail length. Here, we introduce PAL-AI, an integrated neural network machine-learning model that accurately predicts tail-length changes in maturing oocytes of frogs and mammals. We show that PAL-AI learned known and previously unknown sequence elements and their contextual features that control poly(A)-tail length, enabling it to predict tail-length changes resulting from 3'-UTR single-nucleotide substitutions. It also predicted tail-length-mediated translational changes, allowing us to nominate genes important for oocyte maturation. When comparing predicted tail-length changes in human oocytes with genomic datasets of the All of Us Research Program and gnomAD we found that genetic variants predicted to disrupt tail lengthening are under negative selection in the human population, thereby linking mRNA tail lengthening to human female fertility. Overall design: Sythetic mRNA reporter libraries with random and mutagenesis 3'-UTR sequences were injected into frog oocytes. Poly(A)-tail lengths were measured to at different developmental stages to examine the concordance between experimentally determined effects and PAL-AI predicted effects.