Genome annotations for ORF prediction

Genome annotations are constantly evolving, inserting novel transcripts, discarding others and updating annotations of coding sequence. Although most changes are minor from one release to the next, they do accumulate over time (3,33). The initial release of OpenProt was based on genome annotations from December 2015 for each species (24). These were updated to the latest release for each species as of January 2019. This major update necessitated an entire re-run of the OpenProt pipeline (Supplementary Material S1). For example, in humans alone 9873 messenger RNAs (mRNAs) and 2520 non-coding RNAs (ncRNAs) from the NCBI RefSeq annotation, and 1087 mRNAs and 371 ncRNAs from the Ensembl annotation were discarded or had their sequence changed. In parallel, 13 833 novel mRNAs and 2146 novel ncRNAs were included in the latest release of NCBI RefSeq, and 5970 novel mRNAs and 4884 novel ncRNAs in that of Ensembl. Consequently, 4256 RefProts from the NCBI RefSeq annotation and 2785 for Ensembl were discarded or had their amino acid sequence changed, while 4428 RefProts were added from NCBI RefSeq and 7182 from Ensembl. Genome assemblies, annotation releases and the associated number of predicted coding sequences for the OpenProt update are listed in Table 1.

Table 1.

OpenProt (v1.6) prediction pipeline output

Species	Genome assembly	Annotations		ORFeome (both annotations)
		NCBI RefSeq	Ensembl	Total #	Ref #	II_ #	IP_ #
Homo sapiens	GRCh38.p12	GRCh38.p12	GRCh38.95	692 045	134 477	68 612	488 956
Pan troglodytes	Pan_tro_3.0	Pan_tro_3.0	Pan_tro_3.0.95	331 247	79 070	14 308	237 869
Mus musculus	GRCm38.p6	GRCm38.p6	GRCm38.95	558 632	87 339	40 870	430 423
Rattus norvegicus	Rnor_6.0	Rnor_6.0	Rnor_6.0.95	294 727	51 662	7872	235 193
Bos taurus	ARS-UCD1.2	ARS-UCD1.2	ARS-UCD1.2.95	285 565	67 753	11 382	206 430
Ovis aries	Oar_v3.1	Oar_v3.1	Oar_v3.1.95	162 972	30 283	6339	126 350
Danio rerio	GRCz11	GRCz11	GRCz11.95	287 990	68 272	11 896	207 822
Drosophila melanogaster	Release 6 plus ISO1 MT	BDGP6	BDGP6.95	97 834	22 058	2125	73 651
Caenorhabditis elegans	WBcel235	WBcel235	WBcel235.95	94 890	28 516	3034	63 340
Saccharomyces cerevisiae S288c	R64-1-1	R64	R64-1-1.95	16 873	6615	28	10 230

Ref = currently annotated protein (RefProt); II_ = novel isoforms of known protein; IP_ = novel protein from alternative ORF (AltProt).

Interestingly, out of the newly included RefProts, 90 in NCBI RefSeq and 1835 in Ensembl had a length of ≤100 codons. These numbers reflect the efforts from the small ORF and alternative ORF community to foster annotation of these functional yet overlooked proteins (16,17,24). Thus, in humans 736 novel isoforms and 69 AltProts annotated in the OpenProt initial release are now RefProts since they have been included in annotations (NCBI RefSeq, Ensembl and/or UniProt). For example, the protein previously annotated IP_211724 in the initial release of OpenProt (https://openprot.org/p/savedSearch/jCa) was detected and described (renamed altDDIT3) in a publication building on OpenProt data (7). Consequently, the protein was included in the Ensembl annotation (ENSP00000494177) and in UniProt (A0A2R8YD15) in 2018, and became a RefProt in the latest OpenProt release (https://openprot.org/p/savedSearch/kCa). The numbers of AltProts and novel isoforms that have been included in NCBI RefSeq, Ensembl and/or UniProt since the initial release of OpenProt, across all supported species, are listed in Table 2.

Table 2.

OpenProt fosters annotation of novel isoforms and novel proteins

Species	Predicted AltProt (OP v1.3), now RefProt (OP v1.6) (#)	Predicted novel isoform (OP v1.3), now RefProt (OP v1.6) (#)	Total changes (#)
H. sapiens	69	736	805
P. troglodytes	268	1702	1970
M. musculus	28	150	178
R. norvegicus	0	0	0
B. taurus	98	269	367
O. aries	0	0	0
D. rerio	155	661	816
D. melanogaster	1	1	2
C. elegans	8	51	59
S. cerevisiae S288c	0	0	0

OP = OpenProt; v1.3 = OpenProt version 1.3 [annotations from December 2015 (24)]; v1.6 = current OpenProt version (annotations from January 2019); RefProt = currently annotated protein; novel isoform = novel isoform of known protein; AltProt = novel protein from an alternative ORF.

ORF calling from ribosome profiling datasets

In its initial release, OpenProt only reported evidence of translation for ORFs starting with an AUG. Yet pervasive translation with alternative start codons is increasingly reported by the community (14–17), with up to 83% of the sORF repository consisting of non-AUG small ORFs (28). The latest release of OpenProt now integrates data-driven annotation of non-AUG initiating codons. The pipeline of analysis for ribosome profiling datasets (Ribo-seq) uses the PRICE algorithm (21), which can accurately predict non-AUG start ORFs (Supplementary Material S1). Briefly, PRICE uses a logistic regression model to accurately predict initiating codons in Ribo-seq data, whether the dataset is enriched or not for initiating ribosomes. PRICE thus reconstitutes a set of codons most likely to yield the observed reads. We observed that when including all codons as possible initiation sites, the P-value of the reported ORF would increase as the overlap with the ORF predicted by OpenProt decreases (Supplementary Material S2). In PRICE, the P-value is the result of a generalized binomial test; thus, it indicates the confidence of that ORF not being attributable to noise. As the length of the ORF decreases, it is expected to be more difficult to distinguish true translation event from noise. In order to only annotate the most confident alternative initiation starts, we set the following filters based on our observation (Supplementary Material S2): any codon as initiating codon, and an overlap of the ORF candidate with the ORF predicted by OpenProt above 70%. Using these new filters, almost half a million (488 231) translation events could be retrieved in human datasets, corresponding to 33 836 ORFs across OpenProt.

Peptide unicity in MS datasets

Unique and recurrent peptide detection in datasets of MS-based proteomics is considered a gold standard evidence of novel protein expression (1,34). However, the characterization of a unique peptide is often dependent on the database used at the time of the analysis. In classical MS analyses, a peptide is considered unique if it matches to one and only one protein. However, if a protein is not included in the database, it will not be taken into account in the evaluation of the unicity of the peptide. In the initial release of OpenProt, we ensured unicity by using an exhaustive proteome derived from both NCBI RefSeq and Ensembl annotations, as well as UniProt (24). Yet, as mentioned earlier, annotations evolve and with that the proteomic landscape. In the latest OpenProt release, we ensured a control of peptide unicity from MS datasets through time and releases. At each OpenProt release, the set of peptides mapped from all MS datasets to proteins is rescanned for protein assignation enforcing the OpenProt assignation rules. (i) When a peptide matches to multiple protein sequences from different genes, it is discarded. (ii) When a peptide matches to multiple protein sequences from the same gene, with at least one of the proteins being a RefProt, the peptide is given to the RefProt(s) only. (iii) When a peptide matches multiple AltProts or novel isoforms but not RefProts, the peptide is assigned to all AltProts and novel isoforms. Using such rules, OpenProt reports MS-based evidence of expression of novel proteins (AltProts and novel isoforms) only if the detected peptide does not match any known protein.

Furthermore, during this control of peptide unicity, the isobaric residues leucine and isoleucine are treated as undistinguishable. For example, the peptide EIGNLISDAMK was called by our SearchGUI/PeptideShaker MS pipeline to support the identification of IP_658154 (HSPD1P7). However, this peptide only differs by its leucine from the canonical protein of the parental gene HSPD1, which contains the peptide EIGNIISDAMK. Thus, this peptide was re-assigned as EIGNIISDAMK to the RefProt (P10809).

MS statistics to guide data interpretation

Based on the OpenProt pipeline and stringent peptide assignation criteria regarding novel proteins, one must realize that not all predicted AltProts or novel isoforms are detectable by MS (35). To help the OpenProt users with the interpretation of an MS score of 0, we added MS coverage statistics to the MS tab of the details page for each protein. The coverage statistics tab contains the number of independent datasets in which the protein was detected, the number of unique peptides detected across all datasets and the number of peptide spectrum matches across all datasets. Alongside these metrics, the theoretical and current coverages by MS for each given protein are indicated (named possible sequence coverage and detected sequence coverage, respectively). The theoretical coverage is calculated from all possible tryptic peptides that would fit the OpenProt criteria to be assigned: a minimal length of seven amino acids, a maximal mass of 4600 Da and peptide unicity given the protein type (RefProt, novel isoform or AltProt). Thus, in H. sapiens 44 350 of OpenProt annotated proteins would not be detectable by MS, 13 595 in P. troglodytes, 31 275 in M. musculus, 11 311 in R. norvegicus, 11 454 in B. taurus, 8592 in O. aries, 15 294 in D. rerio, 1575 in D. melanogaster, 2267 in C. elegans and 245 in S. cerevisiae S288c.

A summary of experimental detections across all species supported by OpenProt is provided in Table 3 and general statistics are now provided in the About page of the OpenProt website (https://openprot.org/p/about; Supplementary Material S3) alongside a link to a stand-alone, freely accessible version of the script for calculating MS statistics.

Table 3.

OpenProt (v1.6) evidence collection output

Species	Conservation evidence				Translation evidence (Ribo-seq)				Protein evidence (MS)
	Sp #	Ref #	II_ #	IP_ #	St #	Ref #	II_ #	IP_ #	St #	Ref #	II_ #	IP_ #
H. sapiens	9	116 295	45 938	217 897	69	31 453	6912	9738	95	120 925	2446	40 883
P. troglodytes	9	73 137	13 315	131 321	0	N/A	N/A	N/A	2	25 821	15	360
M. musculus	9	78 193	25 867	98 047	40	23 740	3309	5908	40	66 875	236	4458
R. norvegicus	9	49 994	6012	77 768	3	16 205	996	1731	8	21 210	16	398
B. taurus	9	64 866	7148	83 993	0	N/A	N/A	N/A	4	20 120	6	80
O. aries	9	30 091	6223	73 034	0	N/A	N/A	N/A	4	8618	22	104
D. rerio	9	58 386	6826	26 669	5	11 186	490	299	8	32 286	27	220
D. melanogaster	9	13 567	757	477	3	5370	147	243	4	11 560	24	168
C. elegans	9	13 182	1064	450	5	12 796	396	196	6	13 332	38	96
S. cerevisiae S288c	9	2895	5	34	6	5429	4	269	6	4191	0	44

Sp = number of species evaluated for orthology relationships (not counting the queried species); St = number of studies re-analysed by OpenProt; Ref = currently annotated CDS (RefORF); II_ = novel isoforms of known CDS; IP_ = novel CDS from alternative ORF (AltORF); N/A = when no dataset has been re-analysed for this species yet (OpenProt release 1.2). Conservation evidence = all proteins with at least one ortholog in at least one species. Translation evidence = all ORFs detect in at least one detection by PRICE analysis of Ribo-seq data. Protein evidence = all proteins with at least one unique peptide in at least one study.

Identification of similarities of coding sequences within genes

In this OpenProt update, the relationships between proteins from a same gene are now reported. In order to predict proteins as isoforms, OpenProt evaluates the protein sequence identity between proteins from a same gene using an all-versus-all BLAST (Basic Local Alignment Search Tool) (36). OpenProt reports proteins as isoforms if the result of their BLAST search yields a bit score over 40 for an overlap over 50% of the queried sequence, as previously published (37,38). The bit score was chosen as it is dependent on the size of the alignment, but not on the size of the database used. Thus, the higher the bit score, the better the sequence similarity. This isoform feature allows the user to grasp the complexity brought by alternative splicing and by the polycistronic nature of a transcript. For example, the AltProt IP_191523 from the PIDD1 gene is related to six other AltProts from different transcripts of the PIDD1 gene. These range from 284 amino acids (IP_191530) to 109 (IP_784984) (Figure 2A). In parallel, the canonical protein of the same transcript (ENSP00000416801) has eight isoforms ranging from 934 amino acids (XP_005253063) to 526 (Q9HB75-5).

Figure 2.

Annotation of novel protein relationships in OpenProt. (A) Content of the isoforms tab from the details page for IP_191523 on OpenProt (v1.6—available here). The protein accession always sits at the top left corner of the page. The isoform tab displayed here is the last tab of the page. The number indicated at the top corresponds to the number of identified isoforms. The tab contains a protein tree where nodes correspond to unique protein sequences present in OpenProt and are coloured based on their identity to the looked-up protein (here, IP_191523). Each node is accompanied by the corresponding protein accession. By clicking on the protein accession, a pop-up window (represented in dotted boxes) appears with the results of the BLAST search. (B) Content of the top of the isoforms tab from the details page for II_794710 on OpenProt (v1.6—available here). The protein accession always sits at the top left corner of the page. In the case of novel isoforms (accession starting with II_), a phrase under the accession indicates the RefProts responsible for the novel isoform annotation (circled in a red dotted line). The top of the isoform tab, displayed here, will thus contain an additional table. This table contains the accession of the RefProts responsible for the isoform annotation of the looked-up protein (here, II_794710), and the associated reason. Please note that because the filter for a novel protein to be annotated as novel isoform is more inclusive than that of the prediction of related proteins from the same gene, the RefProts listed in the table may not appear in the predicted isoform tree.

Furthermore, OpenProt now also displays for novel isoforms (accessions starting with II_) the RefProt(s) responsible for the annotation as isoform instead of AltProt. This information is displayed at the top of the details page for each novel isoform and in detail in the isoforms tab. For example, the novel isoform II_794710 is annotated isoform because of four RefProts sharing identical genomic coordinates (start and/or end) and a sequence identity over 20% of the protein length. These are NP_001310470, NP_071322 and XP_005270110 from NCBI RefSeq, and ENSP00000359149 from Ensembl (Figure 2B).

Database development

All data are generated using in-house Perl (version 5.18.2) and Python (version 3.6.9) scripts and stored in a PostgreSQL database (version 9.6). All re-analysed MS and ribosome profiling studies are accessible from the Help page (https://openprot.org/p/help; Supplementary Material S4).

Summary of the features presented in the initial release

At the time of the initial release, the OpenProt website provided three interfaces: a genome browser, a search page and a platform for downloads. The genome browser enables rapid lookup of all transcripts, proteins and detected peptides for a specific genomic region, while the search interface is designed for exploration of specific genes, transcripts and/or proteins. The platform for downloads allows custom retrieval of any data on the OpenProt resource as tsv, fasta or bed files (24,35). Each predicted ORF and encoded protein annotated in OpenProt has its own page that contains a genome browser centred on the dedicated ORF alongside all supporting information. All data from the initial release are still present on the website (under the release number 1.3); however, as genome annotations evolve, so does OpenProt, and we strongly encourage using the latest OpenProt release.

Novel OpenProt web interface

The OpenProt resource is hosted at www.openprot.org since its initial release. The graphics of the website have, however, considerably changed for a more user-friendly and graphical interface. The home page (https://openprot.org/) now hosts a presentation of the project with tutorials and visual explanations. The query interface has been updated with an additional advanced search filter. The users can now filter results for specific MS or ribosome profiling datasets (Figure 1). Furthermore, a novel interface has been added for the submission of MS or ribosome profiling datasets to be analysed by OpenProt. This novel submission platform allows users to query their datasets for unannotated proteins, independently of their bioinformatics skills or computational resources. To ensure the quality of the data inserted into the OpenProt database, users first have to deposit their data into public repositories, the PRIDE archive (39) or ProteomeXchange (40) for MS datasets and Gene Omnibus (41) for ribosome profiling ones. After submission, the parameters are validated by the members of the OpenProt team prior to the analysis with the OpenProt pipeline (24) (Supplementary Material S1). Upon completion of the analysis, the results are sent to the users and the data are stored to be inserted in the next OpenProt release. For computational reasons, releases will be launched when 10 datasets are ready for implementation, or if it is 60 days since a dataset has been analysed. The Help page and contents have been updated to support this new interface (https://openprot.org/p/help; Supplementary Material S5).

Website development

The OpenProt web platform was built using the Flask framework (version 1.1.2) and developed using HTML, SQL and JavaScript. The OpenProt website uses an HTTPS protocol to ensure protection of personal information. OpenProt is accessible via all major web browsers supporting JavaScript, such as Safari, Firefox, Chrome or Internet Explorer. All pages can be viewed on mobiles, but the interfaces have been optimized for display on computers or tablets.