GAEVAL is a program for computing coverage and integrity scores for gene models using transcript alignments. The integrity score is a value between 0 and 1 that indicates the level of agreement between the gene model and any related transcript alignments, with 0 corresponding to no transcript support and 1 corresponding to complete transcript support.
Input for GAEVAL is two GFF3 files, one containing gene predictions/annotations and one containing transcript alignments. Although the GFF3 Specification explicitly supports several similar encoding conventions for alignment features, only one is supported by GAEVAL, as shown below.
ctg123 . cDNA_match 1050 1500 5.8e-42 + . ID=match00001;Target=cdna0123 12 462 ctg123 . cDNA_match 5000 5500 8.1e-43 + . ID=match00001;Target=cdna0123 463 963 ctg123 . cDNA_match 7000 9000 1.4e-40 + . ID=match00001;Target=cdna0123 964 2964
GAEVAL will accept alignments with types cDNA_match, EST_match, and the generic nucleotide_match, depending on the source of the data.
The Target and Gap attributes are not disallowed by GAEVAL, but they are not interpreted by GAEVAL either. Gapped or spliced alignments must be encoded as multifeatures, with each segment of the alignment on its own distinct line and all segments of a single alignment sharing the same ID attribute.
GAEVAL computes coverage and integrity scores for each mRNA feature in the gene prediction input. The output will be identical to the input, except that each mRNA feature will have two new attribtues: gaeval_coverage and gaeval_integrity.
The calculation of coverage is straightforward: GAEVAL computes the percentage of nucleotides in exons that have coverage from one or more transcript alignments.
The calculation of integrity is a bit more complex. The integrity score for each gene prediction is a composite of 4 values.
- \(A\): the percentage of introns confirmed by an alignment gap; for single-exon gene predictions lacking introns, \(A\) represents the ratio of the observed CDS length to the expected CDS length (with a maximum of 1.0)
- \(B\): the exon coverage
- \(\Gamma\): the ratio of the observed 5’ UTR length to the expected 5’ UTR length (with a maximum of 1.0)
- \(E\): the ratio of the observed 3’ UTR length to the expected 3’ UTR length (with a maximum of 1.0)
A weight is applied to each of these 4 values, and the final integrity score \(\Phi\) is computed as follows.
The sum of the weights must be 1.0, and their default values are as follows.
- \(\alpha = 0.6\)
- \(\beta = 0.3\)
- \(\gamma = 0.05\)
- \(\epsilon = 0.05\)
Expected lengths for UTRs and CDSs should be determined empirically. The original GAEVAL tool calculated these values as the length achieved by 95% of the evaluated features.
For a complete description of GAEVAL’s command-line interface, run the following command (after AEGeAn has been installed).