DASES contains raw data from two sources. The first source includes total RNA sequencing data on both tumor and adjacent normal tissues from 63 ESCC patients in West China Hospital of Sichuan University. The second source includes publicly available total RNA sequencing data on ESCC patients from the European Bioinformatics Institute (EBI). To ensure high-quality data, we employed strict search criteria to select suitable samples from EBI (Figure 1): 1) the samples were obtained from human ESCC tissues; 2) the data included RNA sequencing; and 3) the data had sufficient information available. We excluded the cell line RNA-seq data, RNA-seq data from esophageal adenocarcinoma or other parts of the esophagus, and any data without sufficient information. It is essential to emphasize that the included datasets were not specifically targeted or enriched for circular RNA (circRNA) or small RNA during the sequencing and library preparation processes.

In this study, we used a series of bioinformatics tools to identify potential lncRNAs and mRNAs associated with ESCC (Figure 2). First, the raw reads obtained from RNA-seq were subjected to quality control using Trim Galore software (version 0.6.4; https://www.bioinformatics.babraham.ac.uk/projects/trim_galore) to obtain clean reads. Then, we used STAR (version 2.7.3a) (Dobin et al., 2013) and HISAT2 software programs (version 2.2.1) (Kim et al., 2019) for sequence alignment of the clean data, resulting in the generation of BAM and SAM files for each sample, respectively. Next, we used Cufflinks (version 2.2.0) (Trapnell et al., 2010) and StringTie software programs (version 2.1.4) (Pertea et al., 2015) to assemble the BAM and SAM files, respectively, generating GTF files for each sample. We then used StringTie software to merge all the assembled GTF files, obtaining a preliminary merged GTF file. Subsequently, we utilized GffCompare software (version 0.12.2) (Pertea and Pertea, 2020) and reference transcripts to identify potential lncRNAs and mRNAs. We selected transcripts with class code “i” or “u” as potential lncRNA candidates and those with class code “=,” “c,” or “j” as mRNA candidates. Finally, we predicted the lncRNA candidates using CPAT (version 3.0.2) (Wang et al., 2013) and PLEK software programs (version 1.2) (Li et al., 2014), and selected those predicted as non-coding RNAs by both tools as lncRNA candidates. We merged the lncRNA and mRNA candidates to generate a comprehensive GTF file containing all potential lncRNA and mRNA candidates associated with ESCC. All the analyses were conducted using the human genome hg38 (release 84) reference provided by Ensembl (https://ensembl.org/Homo_sapiens/Info/Index).

ASEs can manifest in different ways, including skipping an exon (SE), including or excluding a mutually exclusive exon (MXE), using alternative 5′ or 3′ splice sites (A5SS or A3SS), or retaining an intron (RI). To determine the occurrence of ASEs, we compared two different transcript isoforms derived from the same gene. Specifically, we performed paired comparisons between tumor and adjacent normal groups. In these comparisons, we assigned the term “included isoform” to the isoform containing exons when comparing two transcripts. Conversely, the isoform lacking exons was referred to as the “excluded isoform.” The designation of the “included isoform” was based on having a shorter intron length, whereas the “excluded isoform” had a longer intron length (Figure 3). By comparing the splice junctions and exon–intron boundaries between these two isoforms, we identified and quantified the specific ASEs present in the transcriptome.

To comprehensively identify ASEs associated with ESCC, we employed rMATS software (version 3.1.0) (Shen et al., 2014) with a stringent splicing difference cutoff of 0.0001. Given the publicly available data literature reports, which indicated that all the whole-transcriptome data utilized dUTP-based library construction techniques, we considered the fr-firststrand library type during the analysis of aligned reads in BAM format. By comparing exon inclusion levels between tumor and adjacent normal groups, we detected differential ESCC-related ASEs. We only retained ASEs with a percent spliced in (PSI) (Katz et al., 2010) value greater than 0 and that were present in at least two samples. The results of splicing with only reads that span splicing junction based on GTF files were selected as the ESCC-related ASEs. Furthermore, to establish the coordinates of ASEs, we considered that each ASE comprises two transcript isoforms, with each isoform potentially containing 0–2 introns. In order to define the boundaries of ASE, we determined the minimum coordinate of the intron within the event as the starting coordinate and the maximum coordinate of the intron as the ending coordinate.

For quantitative analysis of the RNA-seq data, we employed the merged GTF file as the reference annotation. The BAM files, generated from the alignment step, were subjected to subsequent analysis using Cuffnorm (version 2.2.0), which is a part of Cufflinks software. This tool allowed us to estimate the expression levels of individual isoforms, providing fragments per kilobase of transcript per million mapped reads (FPKM) values.

To assess the potential overlap between ASEs and protein domains, we adopted a conservative approach. We focused only on the known protein domains that intersected with ASEs, disregarding the predictions from various tools. Initially, we retrieved protein domain information from the Ensembl database, specifically focusing on the hg38 version of protein domain annotations (release 109), which includes InterPro coordinates, associated transcripts, and corresponding genes. We then mapped the InterPro coordinates onto genomic coordinates using appropriate alignment algorithms as follows: S t a r t g e n o m i c = S t a r t C D S + 3 × S t a r t I n t e r P r o − 3 , E n d g e n o m i c = S t a r t C D S + 3 × E n d I n t e r P r o − 1 , where S t a r t g e n o m i c and E n d g e n o m i c represent the start and end sites of the protein domain on the genome, respectively, S t a r t C D S refers to the first CDS start site of the corresponding transcript, and S t a r t I n t e r P r o and E n d I n t e r P r o indicate the start and end sites of the protein domain on InterPro coordinates, respectively. Subsequently, we scrutinized whether there was any intersection or overlap between the genomic coordinates of protein domains and genomic coordinates of ASEs. In the cases where a protein domain exhibited any intersection or overlap with an ASE, we deemed it as having a significant overlap with the respective ASE.

DASES is readily accessible through its website at http://www.hxdsjzx.cn/DASES, and no registration or login is required for usage. The current version of DASES was deployed utilizing MySQL (version 8.0.18) (http://www.mysql.com) and operates on a Linux-based Aliyun web server. Server-side scripting was implemented using Tomcat (version 8.0) (http://tomcat.apache.org/) and JAVA (version 1.8) (https://www.oracle.com/technetwork/java/index.html), providing the necessary functionality. The user-friendly web interface of DASES was created using Bootstrap (version 3.3.7) (https://v3.bootcss.com) and jQuery (version 2.1.1) (http://jquery.com) for seamless interaction and enhanced user experience. Genomic visualization capabilities were achieved using JBrowse (http://jbrowse.org) and IGV (https://igv.org), while additional visualizations were facilitated by ECharts (https://echarts.apache.org/zh/index.html). The web interface of DASES comprises various modules, including Home, Search, Browse, Genome Browser, Download, and About, ensuring comprehensive and intuitive access to the platform’s features and information.

Following the application of quality control measures, a total of 14 patients, corresponding to 28 samples, were eliminated from the dataset. Currently, DASES encompasses data from the in-house study and 11 publicly available studies (Table 1), comprising a total of 346 samples, with 185 distinct ESCC patients represented. We identified 257 novel lncRNAs (Figure 4A) and a total of 59,094 ASEs by using a tumor versus adjacent normal strategy, with 31,777 belonging to SE, 7,546 utilizing A5SS, 9,279 utilizing A3SS, 2,653 involving MXE, and 7,839 RI (Figure 4B).

To facilitate easy access and utilization of the database, we designed a user-friendly web interface featuring various modules. The Home page provides users with a concise overview of DASES, accompanied by illustrative diagrams showcasing the five major types of ASEs (Figure 5A). The Search page offers four different search options, namely, gene, transcript, ASE ID, and genomic region, facilitating easy and efficient data retrieval (Figure 5B). The Browse page provides a comprehensive list of all ASE IDs, allowing users to narrow down their queries by applying filters based on the ASE type or study name (Figure 5C). The Genome Browser page enables users to visualize the genomic regions associated with ASEs (Figure 5D). The Download page offers convenient access to essential files, including processed GTF file and ASE-related data, which can be downloaded for further analysis (Figure 5E). Lastly, the About page serves as a valuable resource, providing a detailed pipeline overview of the entire database, along with comprehensive explanations of important interface features, including headers and abbreviations for primary tables, enabling users to fully comprehend and navigate the database with ease.

In DASES, we present a comprehensive search system comprising four dimensions (Figure 5B). The first dimension allows users to conduct searches based on the gene ID or gene name, thereby retrieving pertinent gene information alongside details concerning gene-associated ASEs. By employing the second dimension, users can search using the transcript ID, obtaining transcript-specific information, expression levels across samples, and insights into transcript-associated ASEs. The third dimension facilitates searches based on the ASE ID, yielding ASE-related details, including the exon junction count (EJC), intron junction count (IJC), and PSI values. Finally, the fourth dimension empowers users to search by genomic coordinates, resulting in the retrieval of ASE information specific to designated genomic loci.

The Genome Browser in DASES comprises two distinct sections. The first section facilitates the visualization of all ASEs (Figure 5D). Within the Genome Browser page, users can utilize diverse tracks to filter ASEs based on specific criteria, including ASE types and chromosome numbers. They also have the option to display or conceal tracks associated with ASEs, transcripts, genes, and protein domains. The second section is accessible via the Search page (Figure 5F). When users conduct searches for genes, transcripts, or specific ASEs, the pertinent information is presented visually on the Genome Browser. This seamless integration of search results with the Genome Browser offers users a contextual perspective on the genomic location of these elements.

ASEs have been closely linked to tumorigenesis and cancer progression. To investigate whether the frequency of ASEs exhibits an association with TNM staging in ESCC, we conducted a comprehensive analysis using data from DASES. As shown in Supplementary Figures S1A, D, both the frequency of genes undergoing alternative splicing (AS-gene frequency) and the frequency of ASEs in genes exhibiting AS (ASE frequency) exhibited a substantial increase within ESCC tissues when compared to adjacent normal tissues. Furthermore, our analysis unveiled a significant trend in the correlation between ESCC TNM staging and AS-gene frequency (Supplementary Figures S1B, C), as well as ASE frequency (Supplementary Figures S1E, F). These findings underscore a compelling association between ESCC TNM staging and the frequency of ASEs, suggesting their potential relevance in the context of ESCC progression. The source of DASES facilitates the exploration of these intricate relationships, providing a valuable platform for future research in this field.

The expression of COL6A3 in both bulk esophageal tissue and single esophageal tissue samples exhibited a relatively high level, as evidenced by data obtained from the GTEx website (https://gtexportal.org/home/). Utilizing the search interface of DASES, we specifically queried COL6A3 (Figures 6A, B), leading to the identification of four ASEs, i.e., three SE events and one RI event (Figure 6C). Notably, our findings closely align with the observations reported by Ding, who identified three SE-type ASEs of COL6A3 from 11 samples in their study of ESCC tissues (Ding et al., 2021). Intriguingly, in addition to the ASEs reported by Ding, we discovered an additional RI-type ASE, “DASRI00000001151” (chr2: 237342162–237344349), which was not addressed by Ding. This discrepancy could be attributed to our larger sample size, which enabled us to identify more COL6A3-related ASEs. Importantly, we observed statistically significant differences in the PSI values of these four ASEs between the tumor and adjacent normal tissue groups, further highlighting their potential significance in ESCC (Figure 6D). This robust consistency between our findings and those of Ding provides substantial evidence for the reliability of ESCC-related ASEs documented within DASES, thereby reinforcing their validity through corroboration with findings from other literature reports.