*2.3. Quality Control of Sequencing Data*

All obtained RNA sequences and DNA methylation profiling went through diverse quality checks. The number of reads with assigned feature was calculated with *samtools* in the Unix environment [15]. Samples with less than 1,000,000 assigned reads were not included in the further downstream analysis. Finally, 55 samples from the screening cohort passed the RNA quality control, of them, 34 were DCM patients and 21 were controls; 16 samples from the replication cohort passed the RNA quality filter, of them, 11 were DCM patients and 5 were healthy controls. The overall sequencing depth of the samples in the final cohorts is visualized in the histograms (Figure S5A,B). In addition, the probed methylation sites in that Infinium HumanMethylation 450 K BeadChip kit, known to be possibly influenced by genetic variants, were removed from the following downstream analysis, because the influence of genetic variation on methylation profiling was not in the lens of the present study. In addition, probed methylation sites on X and Y chromosomes and those known to cross-hybridize with non-targeted DNA were dropped. Subsequently, 394,247 qualified probed loci with methylation measurements were included in further analysis.

## *2.4. Data Normalization and Batch E*ff*ect Correction*

Principal component analysis (PCA) was done using package *factoextra* in R programming [16] When analyzing the data characteristics of RNA sequences from both cohorts together, a batch effect could be delineated in the principal component analysis derived from the normalized count matrix. In the PCA plot, samples from both cohorts clustered independently from each other, because the first principal component, responsible for up to 40% of the data variances (Figure S2A), significantly represented the distance between the two cohorts, as could be visualized in the PCA plot (Figure S2B). However, since the relative distribution of the DCM samples and controls samples were consistent in both screening and replication cohorts, and no statistical test of samples across different cohorts were planned to be made, we did not use a cross-cohort batch correction. When performing the statistical tests for differential exon use (DEU) between DCM patients and controls, a normalization of the read count for each gene and each exonic part defined by hg19/GrCh37 was performed. Likewise, all methylation measurements were normalized. The batch effects from different sequencing dates and flow cells were removed.
