Table 2.

Gene Level Accuracy of GenomeScan as a Function of Protein Similarity in DraftGene and FinishGene Datasets

Variable		Similarity category/dataset
		10⁻⁵ > P > 10⁻²⁰		10⁻⁴⁰ > P > 10⁻⁸⁰		10⁻¹²⁰ > P > 10⁻¹⁸⁰
		Draft	Finish	Draft	Finish	Draft	Finish
No. of genes in dataset	174	174	151	151	93	93
% of fragmented genes	42	0	43	0	55	0
No. of predicted genes	186	172	205	159	152	104
Genes completely covered (%)	38	58	48	71	57	73
Genes partially covered (%)	49	32	51	28	42	27
Genes missed (%)	13	10	1	1	1	0
No. of “extra” predicted genes	18	14	19	10	8	11

Sequences were grouped according to the level of similarity between the encoded protein and the available database proteins used in the predictions as described in the legend to Fig. 3. All known genes in the FinishGene set are complete (all coding exons present in a single sequence). Some genes in the DraftGene set represented by multiple “partial genes” in different draft contigs; these are listed as fragmented genes. Known genes were classified as completely covered if all exons were covered by GenomeScan predicted exons; partially covered, if some exons (but not all) were covered byGenomeScan predicted exons; and missed, if no exon was covered by a GenomeScan-predicted exon. GenomeScan predicted genes which did not overlap any known gene are listed as “extra” predicted genes.
↵Includes predicted partial genes as well as complete genes.

Computational Inference of Homologous Gene Structures in the Human Genome