Analysis of Gencode sequences using MisPred tools

Conflict 1

AC110015.1-002 contains an extracellular Cadherin domain but lacks signal peptide or transmembrane domains. AC110015.1-002 is a short (N-terminally and C-terminally truncated) fragment of CADH2_human (that is why it lacks both the N-terminal secretory signal peptide and transmembrane domain of the protein).

Conflict 2

We have identified no sequence based on conflict 2.

Conflict 3

We have identified no sequence based on conflict 3.

Conflict 4

In order to identify suspicious Gencode sequences (with domains deviating from the ‘normal’ size) we focused on those Pfam-A domains (29,561 domains belonging to 2,752 PfamA families) that are present in human Swiss-Prot proteins (altogether 11,384 proteins) as we deemed Swiss-Prot the most reliable protein database. First, we have determined the "normal size-range" of complete Pfam-A domains predicted in the best-curated Swiss-Prot entries and used these criteria to find domains/proteins that significantly deviate from these values. In order to achieve this we removed incomplete domains: i.e. domains whose beginning or end was missing in their first or last 5 positions in the multiple alignments provided by Pfam for each domain, furthermore the lengths of which differ considerably (by at least 2 standard deviations) from the PfamA seed averages.

In the next step we ran a Blastp search with the Gencode sequences as queries against the 16,465 human Swissprot PfamA domains. If the domain match with an Ensembl protein was considerably shorter (in this conservative approach the cutoff was set to 40% of the actual domain length) or longer than the actual domain the Gencode protein was considered suspicious. With this approach we identified 67 Gencode proteins with domains of suspicious length. The results of these analyses are summarized in Table 10.

It should be noted that most of the Gencode sequences identified by conflict 4 are known to have no start and/or no stop codon (cf. Table 10, 4th column), a reflection of the fact that the corresponding transcripts are not full length. In most cases the deviant domains are N-terminally and/or C-terminally truncated as a consequence of the incompleteness of the transcripts (cf. Table 10, 5th column). In the present set of sequences we identified only three cases where the domain deviates from normal size as a result of internal deletion. Further analysis of these cases suggests that the deletions are incompatible with the viability of the protein, suggesting that the transcripts arise through aberrant splicing.

Table 10.  Domain size deviation in Gencode sequences

Recently we have analyzed genome-wide Gencode data sets with the MisPred tools. Our analyses of 74382 human sequences have assigned 75 proteins to the conflict 1 category (Table 11) and 1038 proteins to the conflict 4 category (Table 12). No protein was assigned to the conflict 2 or conflict 3 categories. Note that the dominance of errors violating rule 4 and the absence of errors violating rules 2 and 3 is in agreement with our conclusions based on a much smaller sample of Gencode sequences (see above).

General comments on entries identified with MisPred as violating rule 1 (i.e. they have extracellular Pfam-A domains but lack secretory signal peptides and transmembrane domains).

There are cases where the sequence of the query is identical with that of the corresponding Swiss-Prot entry. Some of these are false positives, e.g. because the Pfam-A domain is not obligatory extracellular (for example, C1q domain in the case of caprins such as that encoded by ENST00000251071). Sometimes both the Gencode sequence and the corresponding Swiss-Prot entry violate rule 1 (for example, CELA1_HUMAN encoded by ENST00000293636 also lacks a typical secretory signal peptide).

In the vast majority of cases the sequence of the query deviates from that of the corresponding Swiss-Prot entry.

In a subset of such cases the Gencode sequence lacks secretory signal peptide whereas the Swiss-Prot entry has a signal peptide. Frequently the Gencode sequences is N-terminally truncated and may not even start with a Met (for example, TPA_HUMAN encoded by ENST00000270187). In another subset the transcript encodes an abnormal variant that deviates from the normal protein at its N-terminal end (for example, ENST00000293922 of C16orf38-201). In another subset translation is N-terminally extended to an incorrect Met, upstream from the valid initiation site (for example, ENST00000267620 for FBLN5_HUMAN).

In some cases there is no transmembrane region in the Gencode sequence whereas the Swiss-Prot entry has a transmembrane region. Most frequently these Gencode sequences are abnormal variant of type II transmembrane proteins (for example, members of the C-type lectin domain family, members of the tumor necrosis factor ligand superfamily, e.g. transcripts ENST00000315591, ENST00000326176).

General comments on entries identified with MisPred as violating rule 4 (i.e. contain Pfam-A domains that deviate from normal size):

There are cases where the Gencode sequence is identical with the corresponding Swiss-Prot entry. Some of these are false positives: e.g. the domain is ‘split’ by low-complexity region (for example, ENST00000234160 encoding GORS2_HUMAN). In several cases both the Gencode sequence and the corresponding Swiss-Prot entry violate rule 4 because it is not a viable protein (for example, ENST00000185907 encoding ASA2B_HUMAN).

The vast majority of Gencode sequences identified with MisPred tool for Conflict 4 deviate from the corresponding Swiss-Prot sequence due to internal or terminal truncation that truncates Pfam-A domains (for example, ENST00000075430 that encodes a fragment of CTDP1_HUMAN).

Table 11.  Gencode sequences violating rule 1

Table 12.  Gencode sequences violating rule 4