Chromatin and Genomic determinants of alternative splicing

Abstract

Alternative splicing significantly contributes to proteomic diversity and mis-regulation of splicing can cause diseases in human. Although both genomic and chromatin features have been shown to associate with splicing, the mechanisms by which various chromatin marks influence splicing is not clear for the most part. Moreover, it is not known whether the influence of specific genomic features on splicing is potentially modulated by the chromatin context. Here we report a deep neural network (DNN) model for predicting exon inclusion based on comprehensive genomic and chromatin features. Our analysis in three cell lines shows that, while both genomic and chromatin features can predict splicing to varying degrees, genomic features are the primary drivers of splicing, and the predictive power of chromatin features can largely be explained by their correlation with genomic features; chromatin features do not yield substantial independent contribution to splicing predictability. However, our model identified specific interactions between chromatin and genomic features suggesting that the effect of genomic elements may be modulated by chromatin context.

Keywords: Algorithms; Alternative splicing; Chromatin; Deep Neural Network; Exon skipping; Experimentation; Machine Learning; Measurement; Performance; Verification.

Figure 1. Cross-validation prediction accuracy of exon inclusion using chromatin features for three cell types

GM12878 (blood), h1-hESC (human embryonic stem cell) and K562 (leukemia). The accuracy is the mean accuracy of 8-fold cross validation. (A) Prediction accuracy using chromatin features; (B) Prediction accuracy using genomic features.

Figure 2

(A) The effect size of chromatin features at different genome locations in h1-hESC cell line; (B) The relevance of chromatin features at different genome locations in h1-hESC cell line.

Figure 3. Cross-validation prediction accuracy using raw genomics (chromatin respectively) features and chromatin (genomics respectively) feature prediction score as an additional feature, for three cell types, GM12878 (blood), h1-hESC (human embryonic stem cell) and K562 (leukemia)

The accuracy is the mean accuracy of 8-fold cross validation. RG indicates the raw genomics features, PC indicates prediction score using chromatin features. RC indicates raw chromatin features, PG means prediction score using genomics features. (A) Comparison between accuracy using RG + PC and only RG; (B) Comparison between accuracy using RC + PG and only RC.

Figure 4. R-squared for explaining residuals of genomics feature prediction using chromatin features and residuals of chromatin feature prediction using genomics features, in three cell lines, GM12878 (blood), h1-hESC (human embryonic stem cell) and K562 (leukemia)

Chro-res: chromatin feature explain residuals of genomics model. Gen: genomics model. Gen_res: genomics feature explain residuals of chromatin model. Chro: chromatin model. (A) R-squared of Chro-res and Gen; (B) R-squared of Gen-res and Chro.

Figure 5. Potential interactions for chromatins-genomics, chromatins-chromatins in GM12878

The red line means negative to exon exclusion, green line means positive to that. The numbers on the line indicate feature location (Fig. 9).

Figure 6. Potential interactions for chromatins-genomics, chromatins-chromatins in h1-hESC

The red line means negative to exon exclusion, green line means positive to that. The numbers on the line indicate feature location (Fig. 9).

Figure 7. Potential interactions for chromatins-genomics, chromatins-chromatins in K562

The red line means negative to exon exclusion, green line means positive to that. The numbers on the line indicate feature location (Fig. 9).

Figure 8. Distribution of exon inclusion level for three cell lines. X-axis is the exon inclusion level, which is between 0 and 1

(A) Distribution for GM12878; (B) Distribution for h1-hESC; (C) Distribution for K562.

Figure 9. Predictive model for exon inclusion prediction. We extracted features from the 7 regions in yellow in the skipping exon event structure

We employed deep neural network model to perform supervised learning to predict exon inclusion.

Figure 10

The deep neural network architecture we used.