Predicting Splicing from Primary Sequence with Deep Learning

Abstract

Graphical AbstractHighlights d SpliceAI, a 32-layer deep neural network, predicts splicing from a pre-mRNA sequence d 75% of predicted cryptic splice variants validate on RNA-seq d Cryptic splicing may yield 10% of pathogenic variants in neurodevelopmental disorders d Cryptic splice variants frequently give rise to alternative splicing A deep neural network precisely models mRNA splicing from a genomic sequence and accurately predicts noncoding cryptic splice mutations in patients with rare genetic diseases. SUMMARYThe splicing of pre-mRNAs into mature transcripts is remarkable for its precision, but the mechanisms by which the cellular machinery achieves such specificity are incompletely understood. Here, we describe a deep neural network that accurately predicts splice junctions from an arbitrary pre-mRNA transcript sequence, enabling precise prediction of noncoding genetic variants that cause cryptic splicing. Synonymous and intronic mutations with predicted splice-altering consequence validate at a high rate on RNA-seq and are strongly deleterious in the human population. De novo mutations with predicted splice-altering consequence are significantly enriched in patients with autism and intellectual disability compared to healthy controls and validate against RNA-seq in 21 out of 28 of these patients. We estimate that 9%-11% of pathogenic mutations in patients with rare genetic disorders are caused by this previously underappreciated class of disease variation.(legend continued on next page) (F) Relationship between exon-intron length and the strength of the adjoining splice sites, as predicted by SpliceAI-80 nt (local motif score) and SpliceAI-10k. The genome-wide distributions of exon length (yellow) and intron length (pink) are shown in the background. The x axis is in log-scale. (G) A pair of splice acceptor and donor motifs, placed 150 nt apart, are walked along the HMGCR gene. Shown are, at each position, K562 nucleosome signal and the likelihood of the pair forming an exon at that position, as predicted by SpliceAI-10k. The genome-wide Spearman correlation between the two tracks is shown. (H) Average K562 and GM12878 nucleosome signal near private mutations that are predicted by the SpliceAI-10k model to create novel exons in the GTEx cohort.

Predicting the impact of single nucleotide variants on splicing via sequence‐based deep neural networks and genomic features

“…Hence, their integration by machine learning leads to moderate performance in predicting a mutation that causes splicing disruption. This observation is consistent with the report that SpliceAI achieved prediction capability by evaluating the change in splicing site strengh caused by a mutation (Jaganathan et al, ).…”

“…This observation is consistent with the report that SpliceAI achieved prediction capability by evaluating the change in splicing site strengh caused by a mutation (Jaganathan et al, 2019).…”

Future directions for high‐throughput splicing assays in precision medicine

“…Sequence context from the whole pre-mRNA transcript affects the order of intron removal and can lead to alternative splicing events not captured in MinGenes (Kim et al, 2017). Moreover, as splicing is a cotranscriptional process, chromatin binding state, absent in minigenes, has also been shown to affect splicing (Jaganathan et al, 2019). Lastly, large data sets containing genetic variants for screening by MPRAs often lack corresponding phenotypic or other relevant MyCode participants are also consented for recontact for additional research which allows for additional clinical evaluation with more targeted phenotyping to supplement the rich data set that already exists in the EHR.…”

“…These analyses further highlight the utility of MRPAs in not only assessing functionally a variant's effect on splicing but also in the construction of predictive models. In addition to MPRA‐trained prediction models, a recent splicing model was trained on primary sequence alone and produced reliable splicing predictions (Jaganathan et al, ). The advantage of this model is that it can identify long‐range sequence features that are not captured in the minigenes used in MPRAs.…”

Assessing predictions of the impact of variants on splicing in CAGI5

“…Furthermore, both high-throughput splicing assays and prediction methods are undergoing rapid developments (e.g., Jaganathan et al, 2019). Furthermore, both high-throughput splicing assays and prediction methods are undergoing rapid developments (e.g., Jaganathan et al, 2019).…”