How to annotate genome with transcriptomic data

[shedurkin]

Issue Category

Please select the category that best describes your issue:

• Server/Infrastructure (Raven, Klone, Gannet connectivity, performance, or storage issues)
• Bioinformatics Tools (Tool installation, container setup, pipeline errors, workflow issues)
• R/RStudio (Package installation, server connectivity, environment configuration)
• Data Management (File transfers, storage, backup procedures, data organization)
• Access Management (Account requests, permissions, authentication setup)
• Lab Meeting Topic (Suggestions for lab meeting presentations or discussions)
• Equipment/Hardware (Lab equipment issues, hardware problems)
• [ X] Other (Please describe in the issue description)

Issue Description

Please provide a clear description of your issue or request:

I'm working with a genome that is assembled to the chromosome level, but doesn't appear to be annotated (i.e. no .gff available). I can, however, find RNA-seq data for my species on NCBI. Can I use that to manually annotated the genome myself? What tools should I look into for this purpose?

Expected Outcome

What were you trying to accomplish or what result were you expecting?

Use-case is to generate feature-level methylation summaries (GbM, exon methylation, etc.)

Additional Context

Please provide any additional information that might be helpful:

Clarification: My study species is Eunicea tourneforti, but we've been using the genome of a close relative, E. knighti, as reference because there was no published tourneforti genome. However, there are now transcriptomic data available for E. tourneforti.

---

For Lab Members:

• If this is urgent, also post in the Roberts Lab Slack
• Check the Roberts Lab Handbook for existing documentation
• Browse existing GitHub Issues to see if similar issues have been resolved

[sr320]

⸻

1. Conceptual overview

You will be creating gene models using multiple lines of evidence:
1. RNA-seq alignments
→ define exons, splice junctions, UTRs
2. Transcript assemblies
→ reconstruct isoforms
3. Protein homology (optional but strongly recommended)
→ improve gene boundaries and recover lowly expressed genes
4. Ab initio prediction (optional but often helpful)
→ fill in gaps where RNA-seq is missing

These are then combined into a consensus GFF.

This is exactly how Ensembl, NCBI, and most non-model organism genomes are annotated.

⸻

2. Core tools to look into (high-level)

A. RNA-seq alignment (splice-aware)

You likely already know these:
• HISAT2 (very common, fast, good splice modeling)
• STAR (excellent, heavier but very accurate)

Output: BAM files aligned to your genome

⸻

B. Transcript assembly from RNA-seq

This turns alignments into exon/intron models.
• StringTie2 ⭐ (current standard)
• (older: Cufflinks — generally avoid now)

Output: transcript GTF/GFF

You can:
• assemble per sample
• merge across samples to get a unified annotation

⸻

C. Evidence-driven gene prediction / annotation

This is where things come together.

🔹 MAKER2 (gold standard, flexible)
Best choice if you want full control
• Takes:
• genome FASTA
• RNA-seq evidence (GFF from StringTie)
• protein homology (FASTA from related species)
• Can run:
• evidence-only
• evidence + ab initio (AUGUSTUS, SNAP)

Output: polished GFF3

MAKER is extremely common for marine genomes.

⸻

🔹 BRAKER2 (more automated)
Best choice if you want automation
• Uses:
• RNA-seq BAMs
• optional proteins
• Automatically trains AUGUSTUS
• Minimal manual tuning

Tradeoff: less transparent, but very effective.

⸻

D. Functional annotation (after gene models exist)

Structural annotation ≠ functional annotation.

Once you have genes:
• InterProScan (domains, GO terms)
• eggNOG-mapper
• BLASTp vs UniProt/SwissProt

⸻