ORF finding and start site refinement:

This history of phage genomics is full of additional genes being recognized after the initial publication. Leaving a gene out causes the protein sequence to not appear in GenBank. Hence the opportunity for future investigators to match it with a Blast search and recognize it as a conserved protein is denied. On the other hand, there are not serious consequences of false positive gene predictions. Hence we use a fairly thorough prediction scheme to try to avoid leaving out any plausible genes.

Collate results from multiple frame prediction programs utilizing:

Length of reading frame.
Codon usage.
Quality of ribosome binding signal.
Relation to adjacent frames.

Conservation of sequence with identifiable homologues.
Conservation of N terminus with identifiable homologues.
Spacing relative to identifiable promoters, transcription terminators, or other identifiable noncoding features.

Gene identification (what kind of gene is it?):

UTHSCSA bioinformatics environment:

Many of the computations were carried out on computers within the UTHSCSA Bioinformatics Center. An additional entry point for UTHSCSA users to explore the available capabilites can be found in tutorial materials developed for the Dept. Biochemistry Molecular Genetics and Biotechnology Advanced Graduate Course.

Psi-Blast.

Psi-Blast, which is easily done directly at NCBI, goes beyond Blast in the case where one or more homologues is found, but none of them have an identified function. Psi-Blast constructs a key from the first round finds (a PSSM) that learns from the residues in the close homologues more about what residues to allow at each position. It then can be used to search again and see through deeper divergence. Psi-Blast was able to form a PSSM between VpV262 orfG and SIO1 sequences and then see deeper to match a T7 gene.
Identification of VpV262 orfG as maturase/packaging enzyme/terminase.

Focused Blast

If one restricts the database searched, say to only viruses, or to only a particular virus, the signal to noise is greatly improved. But the search is now biased. The reported E value increasingly incorporates the assumption that there is a gene to find in the restricted set.
Note: in the VpV262 paper, we described how to do this directly at NCBI. NCBI has since changed the operation of its site so that this strategy does not work there. It can be executed with a local implementation of the NCBI Blast Toolkit.

Family database searches can be done at NCBI and at various family databases.

These use a library of PSSMs (or related structures) previously constructed from alignments of known protein families. These may be able to see across deep divergence to recognize a gene in the new target virus.

Reverse Psi-Blast.

One can construct family searches on the fly by keying a Psi-Blast search with a suspected distant homologue, and allowing it to form its own PSSM. It may then see across the divergence to pick out a gene in the target virus in later iterations.
This requires a local implementation of Blast and the Blast libraries so that the target virus genes can be incorporated into the library (assuming the target virus is not yet submitted to GenBank).

Focused Reverse Psi-Blast

The library can also be restricted in the revese Psi-Blast strategy to increase sensitivity.

Rank order statistics

When using a severely restricted library, the Blast E value is badly biases and becomes unreliable. The following example describes a different statistic that may be used to validate the results of such a search. It requires a prior hypothesis about which gene in a distant virus is considered to be the likely homologue.
Identification of VpV262 portal and capsid genes.