AVISPA: a web tool for the prediction and analysis of alternative splicing

Abstract

Transcriptome complexity and its relation to numerous diseases underpins the need to predict in silico splice variants and the regulatory elements that affect them. Building upon our recently described splicing code, we developed AVISPA, a Galaxy-based web tool for splicing prediction and analysis. Given an exon and its proximal sequence, the tool predicts whether the exon is alternatively spliced, displays tissue-dependent splicing patterns, and whether it has associated regulatory elements. We assess AVISPA's accuracy on an independent dataset of tissue-dependent exons, and illustrate how the tool can be applied to analyze a gene of interest. AVISPA is available at http://avispa.biociphers.org.

Figure 1

AVISPA’s analysis pipeline. The analysis is composed of the following steps. (1) Query submission: users submit a query composed of either a single exon of interest or an exon triplet that also specifies the up- and downstream exons. (2) Query matching: the submitted query is first matched against internal databases (DB) of known transcripts and alternative exons. If no match is found the query is searched against the reference genome. If the query cannot be matched (red cross) an error is reported. (3) Splicing prediction: a successfully matched query (light blue rectangle) is scored as an alternative cassette exon, followed by scoring for differential splicing in four tissue groups. (4) Splicing analysis: if the query’s predictions pass a user-defined significance threshold a splicing analysis is performed. Analysis includes feature enrichment, effect of in silico motif removal on splicing predictions, and mapping putative regulatory motifs to the genome. A visual summary of both predictions and splicing analysis is produced (right).

Figure 2

Prediction accuracy. (a) Differentiating alternative (n = 11,773) from constitutive (n = 9,638) exons. Detecting which exons are alternative (green) is significantly improved compared to a classifier that uses exon expression measurements from 33 experiments (cyan), and compared to the original classifier trained to detect only tissue-dependent cassette exons (red). Detection of exons that exhibit tissue-dependent splicing changes (blue, n = 659) is much more accurate. Numbers within each legend represent the area under the curve (AUC) (b) Identifying tissue-dependent splicing. Detecting tissue-dependent splicing changes (n = 865) from a random set of non-tissue-dependent exons (n = 4,000) achieves an overall accuracy of 89% AUC (black). Accuracy varies considerably between tissues and for detecting increased inclusion (solid line) or exclusion (dashed) in a tissue (c) Detection accuracy for an independent set of Mbnl1/2-dependent exons [14] (n = 461). Differentiating between Mbnl1/2-dependent exons and constitutive exons achieves 97% AUC. Accuracy in detecting Mbnl1/2-dependent exons from a random set of non-tissue-dependent exons (n = 2,000) is approximately 94% AUC for both brain (blue) and muscle (red).

Figure 3

Analysis of Vegfa exon 6 muscle-dependent inclusion. A subset of the summary page produced by AVISPA is shown. (a) Feature enrichment analysis: the values of the features listed on the left are computed for Vegfa exon 6 and compared against matching feature values in a set of labeled exons. The four sets of exons compared against here are alternative exons ('AS', third column from the left), constitutive exons ('Const', third column from the right), exons differentially included in muscle ('Muscle Inc', second column from the right), and differentially excluded in muscle ('Muscle Exc', right most column). Relative enrichment or depletion of features is indicated using the heat map on the right. Only features with significantly low (blue) and high (red) values are shown here. The genomic region of each feature is indicated by the second from left column using the notation and colors in the top figure. (b) Stacked bar chart (left) of the normalized feature effect (NFE, y-axis) on splicing prediction. Only the top motifs are shown. Motif regions are annotated using the color scheme depicted below. Mapping of the motifs onto the UCSC genome browser is shown on the right. Tracks combining all motifs used by the code model (red), the unbiased motif search [5] (grey scaled), and conservation (blue) are added at the bottom.