# Genome Assembly and Annotation for Non-Model Organisms By the end of this guide you will be able to plan a de novo sequencing project for a species with no reference genome, assemble long-read data into contigs, scaffold those contigs to chromosomes, evaluate the result with BUSCO and N50 statistics, and produce a structural and functional gene annotation. You will need a Linux or macOS workstation or HPC account, at least 64 GB of RAM for small genomes (more for large plant genomes), Conda or Apptainer for tool installation, and your raw sequencing data in [FASTQ format](/knowledge/bioinformatics/fastq-file-format-phred-scores-qc). Falling sequencing costs mean that genome projects are now feasible for smaller labs, but a de novo project still requires knowledge of genome size, methods for extracting high molecular weight DNA, and enough tissue, which matters especially for small organisms [1]. This article walks through the whole pipeline for eukaryotic and prokaryotic genomes, compares the main tools, and flags the mistakes that most often ruin otherwise good data. ## What Makes Non-Model Organism Genomes Different A model organism comes with a curated reference genome, known gene models, repeat libraries, and decades of community benchmarks. A non-model organism gives you none of that. Every decision, from k-mer size to [protein database](/blog/guides/protein-database) choice, has to be made from scratch and justified in your methods section. Three practical consequences follow from this. First, assembly quality depends heavily on sequencing technology. Pure short-read assemblies can capture most of the coding parts of a genome, but they are usually much more fragmented and do not resolve repeat elements or structural variants well. [Long-read sequencing](/knowledge/bioinformatics/long-read-sequencing-technologies-pacbio-and-oxford-nanopore) is the method of choice for high-quality assemblies [1]. Second, you have no external ground truth. When Elysia timida was assembled, its scaffold N50 of 41.8 Mb and contig N50 of 1.92 Mb were roughly 30-fold and fourfold higher than other published sacoglossan assemblies, which is how the authors demonstrated improvement in the absence of a gold standard [2]. You will rely on internal metrics (BUSCO, k-mer completeness, read mapping rates) and comparisons with related species. Third, databases are biased toward well-studied taxa. Functional annotation of organisms such as thraustochytrid protists is complicated by limited reliable reference genomes and poor compatibility with curated protein databases [3]. Expect lower annotation transfer rates for underrepresented lineages and plan validation accordingly. ## Planning the Project Before You Sequence ### Estimate genome size and complexity first ![Calculation of long-read sequencing output from a 196 Mb genome estimate at 30x coverage, giving about 6 Gbp of HiFi bases.](/article-images/illustrations/non-model-organism-genome-assembly-depth-budget.webp) You need an approximate genome size before you can calculate [sequencing depth](/knowledge/diagnostics/molecular/how-to-calculate-sequencing-depth-and-coverage-for-your-ngs-run) or budget flow-cell yield. Two cheap options: 1. Flow cytometry on isolated nuclei, if you have fresh tissue. 2. A k-mer histogram from a small pilot of short reads, using Jellyfish and GenomeScope2. ```bash # Count 21-mers in trimmed Illumina reads jellyfish count -C -m 21 -s 5G -t 16 -o reads.jf \ sample_R1.fastq.gz sample_R2.fastq.gz jellyfish histo -t 16 reads.jf > reads.histo # Fit a model to estimate genome size and heterozygosity genomescope2 -i reads.histo -o gs_out -k 21 ``` GenomeScope2 reports estimated genome length, homozygous and heterozygous coverage, and repeat content directly in its output file `gs_out/genomescope2_results.txt`. A typical output line looks like: ``` Genome Haploid Length: 196,392,637 bp Model Fit: 97.4% Heterozygous Rate: 1.07% ``` The 196 Mb figure matters because it tells you what total coverage you should aim for. A 30x long-read target for a 200 Mb genome means roughly 6 Gbp of HiFi bases, plus overage for filtering. ### DNA quality decides the project Long-read platforms are unforgiving of sheared DNA. The brief guide to establishing genome sequencing for non-model and emerging model organisms emphasizes that genome projects hinge on techniques for extracting high molecular weight DNA and on the amount of DNA obtained from a single individual, especially when dealing with small organisms [1]. In practice this means nuclei-based extraction protocols, wide-bore pipette tips, gentle mixing, and checking fragment distribution on a Femto Pulse or TapeStation before committing expensive flow cells. ### Choose the sequencing mix Recent chromosome-level assemblies converge on the same recipe: long reads for contiguity plus proximity ligation for scaffolding. The forked fungus beetle Bolitotherus cornutus used PacBio HiFi plus Hi-C to produce a 196 Mb assembly with ten major chromosomal scaffolds [4]. The endangered little vermilion flycatcher Pyrocephalus nanus used HiFi for contigs and Hi-C for scaffolding to reach a 1.07 Gb genome with scaffold N50 of 74 Mb and 98.9% of sequence assigned to candidate chromosomes [5]. The Malabar red snapper Lutjanus malabaricus combined HiFi with Hi-C scaffolding to anchor 1.0 Gb assemblies to 24 chromosomes with BUSCO above 99% [6]. Flax genotypes were assembled from Oxford Nanopore MinION reads with Flye, polished, and organized into hybrid scaffolds using Bionano optical map data before anchoring into 15 pseudomolecules [7]. The African yam bean combined Nanopore assembly with Hi-C scaffolding for a chromosome-scale assembly with an N50 of 69.5 Mbp [8]. The takeaway for planning is straightforward. HiFi (PacBio) or high-accuracy Nanopore reads give you contigs. Hi-C, optical maps, or strand-seq gives you chromosomes. Illumina reads add accuracy for polishing and are the standard input for annotation pipelines. RNA-seq from multiple tissues feeds gene prediction. ## Non-Model Organism Genome Assembly Step by Step ![Flowchart from raw reads through QC to hifiasm for HiFi or Flye for Nanopore, then purge, polish and optional Hi-C scaffolding.](/article-images/illustrations/non-model-organism-genome-assembly-assembly-branches.webp) The walkthrough below assumes a diploid animal genome of roughly 200 Mb to 1 Gb, assembled from HiFi reads, with optional Hi-C scaffolding and Illumina polishing. File names are placeholders. Adjust paths, thread counts, and memory flags to your system. ### Step 1. Quality control of raw reads Run FastQC on Illumina data and NanoPlot on Nanopore or HiFi data before anything else. ```bash fastqc *.fastq.gz -o fastqc_out -t 8 # Long-read QC summary nanoplot --fastq hifi_reads.fastq.gz --outdir nanoplot_out \ --N50 --plots violin hex ``` Expected output: FastQC opens as one HTML report per file. For good HiFi data, NanoPlot reports a modal read length in the 15 to 25 kb range and a median read quality of Q30 or higher. If your N50 read length is under 5 kb, stop and revisit extraction rather than wasting compute on a poor assembly. ### Step 2. Filter and trim For HiFi reads, remove reads below the platform quality threshold. For Nanopore, filter by quality and length. ```bash # HiFi: keep Q20+ reads filtlong --min_length 1000 --keep_percent 90 \ hifi_reads.fastq.gz > hifi_filtered.fastq.gz # Nanopore: quality and length filter filtlong --min_length 5000 --min_q 10 \ ont_reads.fastq.gz > ont_filtered.fastq.gz ``` ### Step 3. Assemble with a long-read assembler The main contig-level assemblers for long reads are hifiasm and Flye for HiFi, and Flye, Shasta, Raven, wtdbg2, NECAT, and NextDenovo for Nanopore. A benchmark of eleven long-read assemblers on bacterial genomes evaluated runtime, N50, total length, contig count, GC content, and BUSCO completeness, and found that assemblers using progressive error correction with consensus refinement, notably NextDenovo and NECAT, consistently generated near-complete, single-contig assemblies with low misassemblies and stable performance across [preprocessing](/blog/data-analysis/what-is-data-preprocessing) types. Flye offered a strong balance of accuracy and contiguity, although it was sensitive to error-corrected input [9]. For HiFi data, hifiasm is the standard first choice: ```bash hifiasm -o beetle.asm -t 48 hifi_filtered.fastq.gz # Primary contigs awk '/^S/{print ">"$2; for(i=6;i<=NF;i++) print $i}' \ beetle.asm.bp.hap1.p_ctg.gfa > hap1.p_ctg.fasta ``` For Nanopore data, Flye is a reliable default: ```bash flye --nano-hq ont_filtered.fastq.gz -o flye_out \ -t 48 --threads 48 --genome-size 1.0g ``` Key parameters explained: - `-genome-size` (Flye) tells the assembler the expected assembly size, which improves repeat resolution estimates. Use the GenomeScope2 estimate. - `-t` sets threads. Match this to your allocated cores. - hifiasm writes haplotype-resolved outputs. The `p_ctg` (primary) FASTA is the usual starting point for a diploid reference. ### Step 4. Evaluate contig assembly before scaffolding Never scaffold an assembly you have not QC'd. Two tools do most of the work: ```bash # Contiguity and size statistics quast.py flye_out/assembly.fasta -o quast_contigs # Completeness against a related clade's ortholog set busco -i assembly.fasta -l hymenoptera_odb10 \ -m genome -o busco_contigs --cpu 24 ``` Expected output for a decent 700 Mb insect assembly, modeled on the ambush bug Phymata mystica, which assembled to 710 Mb with 99.7% of sequence in 14 chromosomal scaffolds and 58.85% repetitive sequence [10]: ``` # BUSCO results (insecta_odb10) Complete (single + duplicated) 96.8% Fragmented 1.2% Missing 2.0% ``` QUAST reports N50, the contig length at which half the assembly is contained in contigs of that length or longer. For context, published non-model assemblies with HiFi data report contig N50s of 17.8 Mb in a 1.07 Gb bird genome [5], 1.92 Mb in the 41.8 Mb-scaffold sea slug assembly [2], and roughly 42 Mb in 1.0 Gb fish assemblies [6]. Short-read-only or Nanopore-only assemblies without polishing typically land orders of magnitude lower. ### Step 5. Purge duplicates and polish Diploid assemblies often contain duplicated haplotigs representing alternate alleles. purge_dups identifies and removes them based on read depth. ```bash # Map HiFi reads back to the assembly minimap2 -ax map-hifi assembly.fasta hifi_filtered.fastq.gz \ | samtools sort -o aln.bam samtools index aln.bam # Calculate read depth histogram and purge purge_dups/bin/calcuts aln.bam > cutoffs.txt purge_dups/bin/purge_dups -c cutoffs.txt \ -d dups.bed assembly.fasta > purged.fasta ``` Then polish the purged assembly. HiFi assemblies usually need no external polishing, which is one of their main advantages. Nanopore assemblies need Illumina polishing with tools such as Pilon or Racon plus Medaka. ```bash # Racon (Nanopore) then Pilon (Illumina) racon -t 24 ont_filtered.fastq.gz ont_overlap.paf assembly.fasta \ > racon1.fasta minimap2 -ax sr assembly.fasta illumina_R1.fastq.gz illumina_R2.fastq.gz \ | samtools sort -o ill.bam pilon --genome assembly.fasta --frags ill.bam \ --output pilon1 --changes ``` Check the `pilon1.changes` file. Thousands of small corrections in a Nanopore assembly are normal. Tens of thousands in a HiFi assembly signal a problem upstream. ### Step 6. Scaffold to chromosomes with Hi-C Hi-C libraries capture physical proximity of DNA in the nucleus. Reads that map far apart in the assembly but pair together indicate the same chromosome. ```bash # Map Hi-C pairs and scaffold (YaHS example) minimap2 -ax sr assembly.fasta hic_R1.fastq.gz hic_R2.fastq.gz \ | samtools sort -@ 16 -o hic.bam yaha -h -o juicer -x hic.bam # generate contact map files (YaHS pre-step) yhs -a assembly.fasta -o yahs_out -t 48 ``` Validate the result visually in Juicebox by loading the .hic contact map. Each chromosome should appear as a bright diagonal block with visible intra-chromosomal structure. Misjoins appear as off-diagonal streaks that you can manually break and rejoin inside Juicebox and export back to FASTA. As a benchmark for what to expect, the Pyrocephalus nanus assembly placed 98.9% of sequence on candidate chromosomal scaffolds with 267 contigs collapsed into 152 scaffolds [5], and Bolitotherus cornutus reached ten major chromosomal scaffolds matching its karyotype [4]. ### Step 7. Final assembly QC Run the full battery on the scaffolded assembly: - BUSCO at genome mode against the correct lineage dataset. - QUAST for N50, total length, and GC content. - Merqury to estimate consensus quality (QV) and k-mer completeness without a reference. - Read mapping rate of HiFi reads back to the final assembly, which should exceed 95%. ```bash merqury.sh hifi_filtered.fastq.gz assembly.final.fasta -t 24 mq_out ``` Merqury writes `mq_out/*.qv` with the consensus quality value. A HiFi-based assembly commonly reaches QV values above 40, meaning fewer than one error per 10,000 bases. Also run `busco` in genome mode one final time, since scaffolding can occasionally break genes across joins. ## Genome Annotation for Non-Model Organisms Annotation converts an assembly into biology. It has two layers: structural annotation (where the genes are) and functional annotation (what the genes do). Both are harder without a close reference, and best practices for annotation quality in non-model genomes emphasize benchmarking and evidence integration [11]. ### Step 1. Repeat identification and masking Repeats dominate many non-model genomes. Repeats account for almost half of the reef-associated fish Lutjanus malabaricus genome [6] and 58.85% of the Phymata mystica genome [10]. For plants, abundant transposable elements and variable ploidy make this step decisive for gene prediction quality [11]. ```bash # Build a species-specific repeat library, then mask BuildDatabase -name beetle -engine ncbi assembly.final.fasta RepeatModeler -database beetle -engine ncbi -pa 24 RepeatMasker -lib beetle-families.fa -pa 24 \ -gff -dir rm_out assembly.final.fasta ``` The outputs `rm_out/assembly.final.fasta.masked` and the GFF file feed directly into gene prediction. Always build a de novo library with RepeatModeler or EDTA rather than masking only with Dfam or Repbase, because non-model taxa carry repeat families absent from curated libraries. ### Step 2. Gather transcript evidence RNA-seq is the single strongest input for structural annotation. A benchmark across five plant genomes examined the impact of repeat masking, short- and long-read input, and de novo versus genome-guided protein evidence within BRAKER and MAKER workflows, and found that transcripts derived from RNA-read alignments alone are not sufficient for genome annotation. Workflows that combine evidence-based and ab initio approaches were recommended, and combining short and long reads improved annotation. Adding protein evidence from de novo transcriptome assemblies or full-length proteins from related species also helped [11]. Practical protocol: 1. [Sequence RNA](/knowledge/molecular-biology/sequence-rna) from multiple tissues (brain, gonad, gut, whole juvenile) to maximize gene discovery. 2. For Illumina RNA-seq, run Hisat2 or STAR to align reads to the genome, then StringTie for transcript structures. 3. For long-read [transcriptomics](/knowledge/bioinformatics/modern-transcriptomics-bulk-single-cell-spatial), run minimap2 with the `-ax splice` preset for cDNA or direct RNA reads, which recovers full-length isoforms. 4. Assemble proteins from related species into a homology set. A protocol guide for RNA-seq analysis in model and non-model organisms provides step-by-step workflows from raw reads to functional enrichment, with scripts and a sample dataset in a public repository, and discusses tool choices for the same task [12]. If you have no genome at all and need a transcriptome, de novo assembly remains essential. Recent benchmarks of long-read [de novo transcriptome assembly](/knowledge/bioinformatics/transcriptome-assembly-without-a-reference-genome) tools (RATTLE, RNA-Bloom2, isONform) against the short-read assembler Trinity confirmed that long reads generate longer assembled transcripts than short reads for reference-free analysis, though limitations remain compared to reference-guided approaches [13]. A structured framework for de novo transcriptome reconstruction in non-model systems recommends comparing short-read, long-read, and hybrid assembly approaches and layering homology-based, domain-based, and ontology-driven annotation [14]. A worked non-model example is the thraustochytrid protist Ulkenia visurgensis, assembled entirely with free software, which yielded 45,867 unique gene models, 66,623 transcripts, and a contig N50 of 3,162 bp [3]. ### Step 3. Structural gene prediction For eukaryotes, the two dominant frameworks are BRAKER and MAKER. The plant benchmark compared both across five genomes and found that results depended on genome quality, genome complexity, sequence read input, and method, so structural benchmarks that reflect gene structures, reciprocal similarity searches, and mono-exonic versus multi-exonic gene counts give a more complete view of annotation accuracy than BUSCO alone [11]. A BRAKER3 run integrating proteins and RNA-seq alignments: ```bash braker.pl --genome assembly.final.fasta \ --prot related_species.proteins.fasta \ --rnaseq hisat2_alignments.bam \ --species=fly --softmasking --gff3 --AUGUSTUS_ab_initio \ --workingdir braker_out ``` Key parameters: - `-softmasking` tells the pipeline that lowercase bases are soft-masked repeats, so gene predictors avoid calling genes inside repeats. - `-prot` supplies protein homology from related species. Keep this set close phylogenetically. Adding distant homologs increases false positives. - `-rnaseq` supplies spliced alignments as intron evidence. For a modular, evidence-integrated alternative, a pipeline built for the flatworm Schmidtea mediterranea combines Illumina and PacBio sequencing, reference-based and de novo transcript assembly, deep-learning splice site prediction with DeepSplice, and functional filtering of low-confidence transcripts. Applied to the asexual strain it revealed thousands of previously undescribed putative genes and improved existing gene models, and the source code is publicly available [15]. After prediction, filter suspicious models: remove genes entirely inside repeats, discard single-exon models with no protein or RNA support unless your lineage has many genuinely single-exon genes, and use the `getnonredun.pl`-style deduplication or remove transcripts that overlap identically on the same strand. ### Step 4. Functional annotation Functional annotation transfers names and functions onto predicted proteins. ```bash # Orthology-based annotation emapper.py -i braker_out/augustus.haa/aa_clean.fasta \ -m mmseqs --cpu 24 --output beetle_eggnog # Domain annotation interproscan.sh -i braker_out/proteins.fasta \ -f TSV -goterms -pa # Similarity search against curated databases blastp -query proteins.fasta -db uniprot_sprot \ -outfmt "6 qseqid sseqid pident length evalue" \ -evalue 1e-5 -max_target_seqs 1 -num_threads 24 > blast.tsv ``` For prokaryotes, the choice of annotator matters and is now evidence-based. A large-scale benchmark of four open-source annotation tools across 156,033 genomes found that Bakta excels at annotating high-quality bacterial genomes, while PGAP performed better on archaeal genomes and on challenging bacterial assemblies including metagenome-assembled, fragmented, or contaminated samples. For Gene Ontology coverage, PGAP consistently provided broader term coverage, whereas EggNOG-mapper offered more terms per feature [16]. ```bash # Bacterial isolate genome bakta --db /path/bakta_db --output bakta_out \ --threads 24 genome.fasta ``` Automatic annotation still requires manual curation, and web tools such as ReNoteWeb exist specifically to assist the curation process by identifying missing gene products through a simple interface [17]. ### Step 5. Validate the annotation - Run BUSCO in protein mode on your predicted proteome. The B. cornutus annotation of 12,459 protein-coding genes [4] and the 19,904 protein-coding genes of E. timida [2] were both assessed against BUSCO completeness relative to related taxa. - Compare gene count to genome size and lineage expectations. The fish assemblies with roughly 25,000 protein-coding genes [6] and the insect with 26,760 [10] sit in familiar vertebrate and insect ranges. - Check that exon counts and intron length distributions match related, well-annotated species. Massive numbers of single-exon genes usually mean unmasked repeats or contaminant. - BLAST a random sample of 50 predicted proteins against UniProt. If most hit bacteria or fungi in an animal genome, hunt for contamination. ## Bioinformatics Tools Comparison | Task | Leading tools | Strengths | Watch out for | |--|--|--|--| | Long-read contig assembly (HiFi) | hifiasm, Flye | haplotype-resolved output, high contiguity | hifiasm memory use on large genomes | | Long-read contig assembly (ONT) | Flye, NextDenovo, NECAT, Raven, Shasta, wtdbg2 | NextDenovo and NECAT produced near-complete, low-misassemblies assemblies in benchmarking [9] | Flye was sensitive to error-corrected input [9] | | Eukaryotic structural annotation | BRAKER3, MAKER | evidence-integrated, well benchmarked [11] | performance varies with genome quality and input type [11] | | Transcriptome (no genome) | Trinity, RNA-Bloom2, isONform, RATTLE | long-read tools produce longer transcripts [13] | isoform resolution still limited [14] | | Prokaryotic annotation | Bakta, PGAP, Prokka, EggNOG-mapper | Bakta best for high-quality bacterial genomes, PGAP better for archaea and fragmented assemblies [16] | tool choice depends on genome quality and taxonomy [16] | | Repeat masking | RepeatModeler2, RepeatMasker, EDTA | species-specific libraries | curated libraries miss lineage-specific repeats | | Scaffolding | YaHS, SALSA, Juicebox (curation), Bionano | chromosome-scale pseudomolecules [3][11] | misjoins need visual curation | | Assembly QC | BUSCO, QUAST, Merqury | completeness, contiguity, consensus QV | BUSCO alone does not measure structural annotation accuracy [11] | This is a practical bioinformatics tools comparison, not an exhaustive catalog. The best bioinformatics software for your project is the combination that matches your sequencing chemistry, genome size, and taxon, validated on your own data. ## Non-Model Organism Genome Assembly Tips 1. Sequence multiple individuals for RNA-seq across tissues, but assemble the genome from one high-quality individual when possible. Mixing individuals inflates apparent heterozygosity and fragments the assembly. 2. Estimate genome size with GenomeScope2 before ordering flow cells so your coverage target is real, not a guess. 3. Reserve Illumina reads for polishing and annotation even in a long-read project. The plant benchmark showed that a combination of short and long reads improves gene prediction [11]. 4. Include long-read transcript data when you can. Nanopore RNA-seq supported gene prediction in the African yam bean annotation [8], and long-read transcriptomics is an emerging development improving transcriptome completeness [14]. 5. Screen for contamination early with FastQ Screen or Kraken2. Thraustochytrid and other environmental isolates routinely carry bacterial symbionts [3]. 6. Keep every intermediate on disk with a log file for each command. Journaling the exact parameters turns a chaotic exploratory project into a reproducible pipeline. 7. Use the correct BUSCO lineage database. Running the insecta dataset on a mollusk assembly produces misleadingly poor numbers. 8. Assemble the [mitochondrial genome](/blog/guides/mitochondrial-genome) separately. The HiFiMiTie pipeline fully assembled and verified all portions of the flycatcher mitogenome from HiFi reads, yielding 17,151 bases with 13 protein-coding genes, 22 tRNAs, 2 rRNAs, and duplicated control regions [5]. ## Common Mistakes and Limitations These are the errors that most often waste data or produce misleading assemblies in non-model organism genome assembly projects. - Skipping genome size estimation. Without it you cannot tell whether low coverage or a truly huge genome caused a fragmented assembly. - Using short reads alone and expecting contiguity. Pure short-read assemblies are much more fragmented and resolve repeats and structural variants poorly [1]. They are a fallback, not a default. - Ignoring heterozygosity. Highly heterozygous outbred individuals produce tangled assemblies. Choosing the least heterozygous, most inbred-feasible individual, or using haplotype-resolved assemblers, mitigates this. - Masking repeats with only curated libraries. Non-model taxa carry repeat families absent from Dfam and Repbase, which causes false gene predictions inside repeats. Build a de novo library. - Treating RNA read alignments as sufficient for annotation. The plant benchmark showed RNA-read alignments alone are not sufficient for genome annotation [11]. - Chasing chromosome-level quality at any cost. Genome initiatives produce standards for sequencing and assembly, but if your organism is not part of such an initiative or does not meet its standards, a genome of low contiguity in intergenic parts that still carries most coding sequence can support publication. Missing chromosome-scale standards should not prevent the work from being shared [1]. - Over-interpreting functional annotations. Database biases and incomplete annotations are recognized limitations, particularly in underrepresented taxa [14]. Labels transferred from distant homologs are hypotheses, not facts. - Trusting BUSCO alone. BUSCO measures ortholog completeness, not structural correctness of gene models. The plant benchmark showed that benchmarks reflecting gene structure, reciprocal similarity alignments, and exon-count distributions give a more complete view [11]. - Forgetting downstream users. A genome without a documented pipeline, QC numbers, and deposited raw reads cannot be independently evaluated. Recent non-model genome papers routinely report contig and scaffold N50, BUSCO percentages, repeat fraction, and gene count together [3][6][11][13][17][18], and that is the standard to match. One further limitation deserves plain statement. Even excellent assemblies have systematic weaknesses: haplotype phasing choices affect apparent genome size, long-read callers still misresolve some repeats, and annotation error rates in non-model taxa remain higher than in model organisms because the training data behind the tools comes from model species [11]. Individual projects need pilot experiments and biological judgment, and there is no substitute for validating a subset of genes by eye. ## Reproducibility, Publication, and Data Sharing Deposit raw reads, the final assembly, and the annotation in NCBI or a recognized archive before submission, and report the metrics reviewers expect: total length, contig and scaffold N50, BUSCO completeness with the lineage dataset named, consensus QV from Merqury, repeat fraction, and predicted gene count. Building a public database alongside the genome, as was done for two hybrid rice restorer lines with search and alignment tools for breeders [18], is one model for making the resource usable. For expression studies that will use your new reference, note that improved assemblies plus single-cell RNA-seq methods are expanding what [gene expression](/blog/guides/gene-expression) studies in non-model organisms can resolve, including disentangling differential [gene regulation](/blog/guides/gene-regulation) from cell type composition [19]. Package your pipeline with Conda environment files or containers, version-control your scripts, and pin tool versions in your methods section. A reviewer who can rerun your numbers is a reviewer who can trust them. ## Frequently Asked Questions ### How much sequencing depth do I need for a non-model genome assembly? Aim for roughly 25x to 35x coverage in long reads for the contig assembly, plus about 30x Illumina for polishing and 50x to 100x Hi-C. The exact base count depends on your genome size estimate from GenomeScope2 or flow cytometry, so multiply the estimated haploid genome size by the target coverage to get the required output. Projects that report their metrics, such as the 1.07 Gb flycatcher assembly, achieved chromosome-scale results from this style of combined design [5]. ### Can I assemble a non-model genome from Illumina short reads only? Yes, but expect a fragmented result. Pure short-read assemblies can capture most of the coding parts of a genome, yet they are usually much more fragmented and do not resolve repeat elements or structural variants well [1]. Use short-read-only assembly when long-read sequencing is genuinely impossible, and consider supplementing it with de novo transcriptome assembly to recover gene structure [14]. ### Which assembler should I choose for my genome? Match the assembler to your chemistry and genome size. For HiFi data, hifiasm is the common first choice, and Flye is a strong default for Nanopore data. Across eleven benchmarked long-read assemblers, those with progressive error correction and consensus refinement, notably NextDenovo and NECAT, produced near-complete, single-contig bacterial assemblies with low misassemblies, while Flye balanced accuracy and contiguity well [9]. Benchmark two assemblers on a subset of your reads before committing. ### Do I need Hi-C scaffolding, or are contigs enough? Contigs are enough if your scientific question tolerates low contiguity in intergenic regions, and missing chromosome-scale standards should not prevent publication in that case [1]. Hi-C is the route to chromosomal scaffolds, as demonstrated in the beetle with ten major chromosomal scaffolds [4], the flycatcher with 98.9% of sequence on candidate chromosomes [5], and the snapper anchored to 24 chromosomes [6]. ### What is a good N50 and BUSCO score for a non-model assembly? There is no single universal threshold, because genome size and repeat content differ enormously across lineages. Use published assemblies from related species as your yardstick. For orientation, recent HiFi plus Hi-C non-model genomes report contig N50s from 1.92 Mb in a sea slug [2] to about 42 Mb in fish [6] and 74 Mb in a bird scaffold N50 [5], with BUSCO completeness above 99% in the fish assembly [6]. ### How do I annotate genes when no reference genome exists for my species or clade? Combine three evidence types: RNA-seq alignments and transcript assemblies from your own species, full-length proteins from related species, and ab initio prediction. Workflows that combine evidence-based and ab initio approaches are recommended, and RNA read alignments alone are not sufficient for genome annotation [11]. Layer orthology-based and domain-based functional annotation afterward, following frameworks designed for non-model systems [2][10]. ### Why do my predicted gene counts look too high or too low? Too many genes usually means unmasked repeats, contamination, or duplicate haplotigs that survived purging. Too few genes usually means insufficient transcript evidence or a BUSCO lineage mismatch during protein evaluation. Check repeat fraction against related species, since repeats can occupy almost half of a fish genome [6] or nearly 59% of an insect genome [10]. ### Can I get a non-model genome published without meeting chromosome-level standards? Yes. If the scientific question can be answered with a genome of low contiguity in intergenic parts, missing the high standards of chromosome-scale assembly should not prevent publication [1]. Report your metrics honestly, deposit the data, and describe exactly how the assembly supports the biology you are claiming. ## Sources 1. [Establishing genome sequencing and assembly for non-model and emerging model organisms: a brief guide.](https://pubmed.ncbi.nlm.nih.gov/40247279/) 2. [Chromosome-level genome assembly of the sacoglossan sea slug Elysia timida (Risso, 1818).](https://pubmed.ncbi.nlm.nih.gov/39375624/) 3. [De novo transcriptome assembly and functional annotation supports potential biotechnological applications for the non-model thraustochytrid Ulkenia visurgensis Lng2.](https://pubmed.ncbi.nlm.nih.gov/40228758/) 4. [A chromosome-level genome assembly of the forked fungus beetle Bolitotherus cornutus, a model system for studying social evolution in the wild.](https://pubmed.ncbi.nlm.nih.gov/40990325/) 5. [Genome Assembly of Pyrocephalus nanus: A Step Toward the Genetic Conservation of the Endangered Little Vermilion Flycatcher of the Galapagos Islands.](https://pubmed.ncbi.nlm.nih.gov/38652799/) 6. [Genome assembly and Full-length transcriptome of Malabar Red Snapper, Lutjanus malabaricus.](https://pubmed.ncbi.nlm.nih.gov/42586981/) 7. [Chromosome-scale assembly of European flax (Linum usitatissimum L.) genotypes and pangenomic analysis provide genomic tools to improve breeding.](https://pubmed.ncbi.nlm.nih.gov/41199203/) 8. [Chromosome-scale assembly of the African yam bean genome.](https://pubmed.ncbi.nlm.nih.gov/39695151/) 9. [Benchmarking long-read assembly tools and preprocessing strategies for bacterial genomes: A case study on E. coli DH5α.](https://pubmed.ncbi.nlm.nih.gov/41141382/) 10. [Genome report: genome sequence of Phymata mystica (Evans), an ambush bug.](https://pubmed.ncbi.nlm.nih.gov/42663046/) 11. [Welcome to the big leaves: Best practices for improving genome annotation in non-model plant genomes.](https://pubmed.ncbi.nlm.nih.gov/37601314/) 12. [RNA-Seq Data Analysis: A Practical Guide for Model and Non-Model Organisms.](https://pubmed.ncbi.nlm.nih.gov/38808970/) 13. [A comprehensive evaluation of long-read de novo transcriptome assembly.](https://pubmed.ncbi.nlm.nih.gov/41709347/) 14. [De Novo transcriptome assembly and functional annotation in non-model organisms.](https://pubmed.ncbi.nlm.nih.gov/42972139/) 15. [A modular pipeline for evidence-integrated genome annotation across species: A case study on Schmidtea mediterranea.](https://pubmed.ncbi.nlm.nih.gov/40930433/) 16. [Large-scale benchmarking of prokaryotic annotation tools across thousands of species.](https://pubmed.ncbi.nlm.nih.gov/42706541/) 17. [ReNoteWeb - Web platform for the improvement of assembly result and annotation of prokaryotic genomes.](https://pubmed.ncbi.nlm.nih.gov/36029977/) 18. [From sequences to prebreeding: Chromosome-level genome assemblies, comparative annotation, and a public database for Minghui 86 and Minhui 3301.](https://pubmed.ncbi.nlm.nih.gov/40840635/) 19. [Rethinking eco-evo studies of gene expression for non-model organisms in the genomic era.](https://pubmed.ncbi.nlm.nih.gov/38721834/)