Metagenome-Assembled Genomes: MetaBat2 + CheckM

Overview

Metagenome-assembled genomes (MAGs) are draft genomes reconstructed from shotgun metagenomic sequencing without culturing individual organisms. The workflow assembles all community DNA together, then bins contigs into putative genomes based on coverage depth and tetranucleotide frequency signatures. MAGs allow you to study the genomic content and metabolic potential of uncultivable microorganisms.

MAG quality tiers (MIMAG standards):

High quality: Completeness ≥90%, Contamination ≤5%, with 23S/16S/5S rRNAs and ≥18 tRNAs
Medium quality: Completeness ≥50%, Contamination ≤10%
Low quality: Completeness <50% and/or Contamination >10%
Only report High and Medium quality MAGs in publications. Low quality MAGs are often artifacts or chimeric.

Dataset

We use the SRA PRJNA46333 human gut microbiome dataset — a well-characterised shotgun metagenomics study of human gut microbial communities with known MAG composition from previous analyses.

Property	Value
Sample type	Human gut microbiome (stool)
Accession	PRJNA46333 — HMP Human Microbiome Project
Sequencing	Paired-end Illumina WGS (shotgun metagenomics)
Samples	3 gut metagenomes (for co-assembly binning)
Expected MAGs	10–30 medium-to-high quality MAGs per sample
Dominant phyla	Firmicutes, Bacteroidetes, Proteobacteria

Directory Structure

mag_pipeline/

mag_pipeline/ ├── 0_raw/ # raw paired-end FASTQ files ├── 1_trimmed/ # QC-trimmed reads ├── 2_assembly/ # MEGAHIT co-assembly output │ └── final.contigs.fa # assembled contigs >1 kb ├── 3_coverage/ # per-contig depth files (jgi_summarize) ├── 4_bins/ # MetaBat2 output bins (one FASTA per bin) ├── 5_checkm/ # CheckM completeness and contamination ├── 6_gtdbtk/ # GTDB-Tk taxonomic classification ├── 7_drep/ # dRep dereplicated MAG set └── 8_annotation/ # Prokka gene annotations per MAG

Step 1 — QC & MEGAHIT Assembly

Bash — Trim reads with fastp, co-assemble with MEGAHIT

######################################################################
## STEP 1: QC trimming and metagenome assembly
## Co-assembly: assemble all samples together
##   Advantage: deeper effective coverage for low-abundance organisms
##   Each organism's reads from multiple samples combine into one assembly
##   This gives better-resolved genomes for rare community members
## MEGAHIT: memory-efficient de Bruijn graph assembler for metagenomes
##   Better than SPAdes for large metagenomes (>50 GB RAM constraint)
######################################################################

mkdir -p 0_raw 1_trimmed 2_assembly 3_coverage 4_bins 5_checkm 6_gtdbtk 7_drep 8_annotation
module load fastp/0.23.4
module load megahit/1.2.9

## Download HMP gut metagenome samples
SAMPLES=(SRR059420 SRR059421 SRR059422)
for SRR in "${SAMPLES[@]}"; do
    fasterq-dump --split-files --outdir 0_raw/ --threads 8 ${SRR}
    gzip 0_raw/${SRR}_1.fastq 0_raw/${SRR}_2.fastq
done

## --- Step 1a: QC trimming with fastp ---
## fastp: faster alternative to Trimmomatic with better adapter detection
## -i/-I         : input R1/R2 files
## -o/-O         : output R1/R2 paired files
## --detect_adapter_for_pe : auto-detect Illumina adapters for PE reads
## --cut_right   : quality trim from 3' end using sliding window
## --cut_right_window_size 5 --cut_right_mean_quality 20 :
##   slide a 5-bp window, trim when average quality drops below Q20
## --length_required 50 : minimum read length after trimming
## --thread 8    : parallel threads
## --json/--html : save QC reports
for SRR in "${SAMPLES[@]}"; do
    fastp \
        -i 0_raw/${SRR}_1.fastq.gz \
        -I 0_raw/${SRR}_2.fastq.gz \
        -o 1_trimmed/${SRR}_R1.fq.gz \
        -O 1_trimmed/${SRR}_R2.fq.gz \
        --detect_adapter_for_pe \
        --cut_right --cut_right_window_size 5 --cut_right_mean_quality 20 \
        --length_required 50 \
        --thread 8 \
        --json 1_trimmed/${SRR}_fastp.json \
        --html 1_trimmed/${SRR}_fastp.html
    echo "Trimmed: ${SRR}"
done

## --- Step 1b: Co-assembly with MEGAHIT ---
## Co-assembly: combine all samples' reads into one assembly
## Build comma-separated lists of all R1 and R2 files
R1_ALL=$(ls 1_trimmed/*_R1.fq.gz | tr '\n' ',' | sed 's/,$//')
R2_ALL=$(ls 1_trimmed/*_R2.fq.gz | tr '\n' ',' | sed 's/,$//')

## megahit: de Bruijn graph assembler optimised for metagenomes
## -1/-2          : paired-end read files (comma-separated for multiple samples)
## --min-contig-len 1000 : only output contigs >= 1 kb
##   Short contigs (<1 kb) are rarely useful for binning and increase noise
##   MetaBat2 requires contigs >= 1500 bp for reliable binning
## --k-min 27 --k-max 127 --k-step 10 :
##   Multi-k assembly: tries k-mers from 27 to 127 in steps of 10
##   Large k: resolves repetitive regions better; small k: better coverage of low-complexity regions
##   MEGAHIT automatically merges results from all k values
## -t 32         : 32 threads
## -m 0.9        : use up to 90% of available memory
## --out-dir     : output directory
megahit \
    -1 ${R1_ALL} \
    -2 ${R2_ALL} \
    --min-contig-len 1000 \
    --k-min 27 \
    --k-max 127 \
    --k-step 10 \
    -t 32 \
    -m 0.9 \
    --out-dir 2_assembly/

## Assembly statistics
echo "Total contigs: $(grep -c '>' 2_assembly/final.contigs.fa)"
echo "Total assembly size (bp):"
grep -v '>' 2_assembly/final.contigs.fa | tr -d '\n' | wc -c

Download:

Assembly strategy decoded

📈Co-assembly vs per-sample	Co-assembly combines reads from all samples before assembly. This gives higher effective coverage for organisms present in multiple samples, producing longer contigs. The trade-off: co-assembly can mix up closely related strains across samples into chimeric contigs. For species-level MAGs: co-assemble. For strain-level resolution: per-sample assembly + cross-mapping coverage is sometimes better.
🎯--min-contig-len 1000	Very short contigs (<1 kb) carry too little information for binning — they cannot be assigned to a genome based on tetranucleotide frequency or coverage alone. MetaBat2 specifically requires contigs ≥1,500 bp (it internally filters shorter contigs anyway). Discarding short contigs also reduces output file size significantly without losing useful genome content.

Step 2 — Coverage Calculation

Bash — Map reads back to assembly, calculate per-contig depth

######################################################################
## STEP 2: Calculate per-contig coverage depth
## MetaBat2 uses BOTH tetranucleotide frequency AND coverage to bin contigs.
## Coverage is the primary signal: contigs from the same organism have
## similar coverage (because the same organism contributes reads in proportion
## to its abundance in the community).
## We map ALL samples back to the co-assembly, then use jgi_summarize_bam_contig_depths
## to compute per-contig, per-sample coverage — this gives MetaBat2 multi-sample coverage,
## which dramatically improves binning quality.
######################################################################

module load bowtie2/2.5.1
module load samtools/1.17
module load metabat2/2.15.0

ASSEMBLY="2_assembly/final.contigs.fa"

## --- Step 2a: Index the assembly for Bowtie2 ---
bowtie2-build --threads 16 ${ASSEMBLY} 3_coverage/assembly_index

## --- Step 2b: Map each sample back to the co-assembly ---
## We map EACH sample separately to get per-sample coverage profiles
## Contigs from the same organism should have correlated coverage across samples
## (if organism X is abundant in sample 2, all its contigs have high coverage in sample 2)
SAMPLES=(SRR059420 SRR059421 SRR059422)

for SRR in "${SAMPLES[@]}"; do
    ## Align reads to the co-assembly
    ## --no-unal  : suppress unaligned reads (saves disk space)
    ## -X 1000    : maximum insert size for valid pairs (1000 bp)
    bowtie2 \
        -x 3_coverage/assembly_index \
        -1 1_trimmed/${SRR}_R1.fq.gz \
        -2 1_trimmed/${SRR}_R2.fq.gz \
        --no-unal \
        -X 1000 \
        -p 16 \
        2> 3_coverage/${SRR}_bowtie2.log | \
      samtools sort \
        -@ 8 \
        -O BAM \
        -o 3_coverage/${SRR}_sorted.bam
    samtools index 3_coverage/${SRR}_sorted.bam
    echo "Mapped: ${SRR}"
done

## --- Step 2c: Summarize contig depths using MetaBat2's helper tool ---
## jgi_summarize_bam_contig_depths: calculates mean and variance of coverage
## for each contig in each BAM file.
## This generates a depth file with one row per contig and one column per sample:
##   contigName  contigLen  totalAvgDepth  sample1.bam  sample1.bam-var  sample2.bam ...
## MetaBat2 uses this depth information directly for binning.
jgi_summarize_bam_contig_depths \
    --outputDepth 3_coverage/depth.txt \
    3_coverage/*_sorted.bam

echo "Depth file created: 3_coverage/depth.txt"
echo "Contigs in depth file: $(tail -n+2 3_coverage/depth.txt | wc -l)"

Download:

🌹

Multi-sample coverage is the key to good binning: When you have 3+ samples, MetaBat2 can identify contigs that co-vary in abundance. If Bifidobacterium longum is present in samples 1 and 3 but not 2, all its contigs will have high coverage in samples 1 and 3 and zero in sample 2. This co-variation pattern is a very strong binning signal — much better than coverage from a single sample.

Step 3 — MetaBat2 Binning

Bash — MetaBat2 automated binning using coverage + tetranucleotide frequency

######################################################################
## STEP 3: Run MetaBat2 to bin contigs into putative MAGs
## MetaBat2 clusters contigs based on:
##   1. Tetranucleotide frequency (TNF): the 4-mer composition of the contig
##      sequence. Each organism has a characteristic TNF signature based on
##      its GC content and codon usage patterns.
##   2. Coverage depth: contigs from the same organism have correlated
##      coverage across samples (same organism = same abundance pattern)
## MetaBat2 uses a Bayesian approach to probabilistically assign contigs to bins.
######################################################################

module load metabat2/2.15.0

ASSEMBLY="2_assembly/final.contigs.fa"
DEPTH="3_coverage/depth.txt"

## --- Run MetaBat2 ---
## -i    : input assembly FASTA file
## -a    : abundance/depth file from jgi_summarize_bam_contig_depths
## -o    : output directory + prefix for bin FASTA files
##         Creates: 4_bins/bin.1.fa, 4_bins/bin.2.fa, etc.
## --minContig 1500 : minimum contig length for binning
##   MetaBat2 requires >= 1500 bp contigs to estimate TNF reliably
##   Shorter contigs are not assigned to bins (go to an "unbinned" set)
## --minClsSize 200000 : minimum bin size in bp (= 200 kb)
##   Bins smaller than 200 kb are likely genome fragments, not complete genomes
##   For fragmented assemblies, lower to 100000 (100 kb)
## --numThreads 32 : parallel threads
## --saveCls : save the bin membership table (which contig → which bin)
## --seed 42 : random seed for reproducibility (MetaBat2 uses random elements)
metabat2 \
    -i ${ASSEMBLY} \
    -a ${DEPTH} \
    -o 4_bins/bin \
    --minContig   1500 \
    --minClsSize  200000 \
    --numThreads  32 \
    --saveCls \
    --seed        42

## --- Count output bins ---
N_BINS=$(ls 4_bins/bin.*.fa 2>/dev/null | wc -l)
echo "MetaBat2 created ${N_BINS} bins"
echo ""
echo "Bin sizes (number of contigs per bin):"
for BIN in 4_bins/bin.*.fa; do
    echo "  $(basename ${BIN}): $(grep -c '>' ${BIN}) contigs"
done

## --- Quick size summary for each bin ---
for BIN in 4_bins/bin.*.fa; do
    BINNAME=$(basename ${BIN} .fa)
    SIZE=$(grep -v '>' ${BIN} | tr -d '\n' | wc -c)
    CONTIGS=$(grep -c '>' ${BIN})
    echo -e "${BINNAME}\t${CONTIGS} contigs\t${SIZE} bp"
done | sort -k3 -rn

Download:

Step 4 — CheckM Quality Assessment

Bash — CheckM lineage workflow to assess completeness and contamination

######################################################################
## STEP 4: Assess MAG quality with CheckM
## CheckM uses a curated set of lineage-specific single-copy marker genes
## to estimate:
##   Completeness:  % of expected marker genes found in the bin
##     (if 90% of Firmicutes marker genes are present = 90% complete)
##   Contamination: % of marker genes found in multiple copies per bin
##     (a single genome should have exactly one copy of each single-copy gene;
##      if many are duplicated, the bin likely contains DNA from 2+ organisms)
##   Strain heterogeneity: % of duplicated markers that are highly similar
##     (similar duplicates = two strains; dissimilar = contamination from another species)
######################################################################

module load checkm/1.2.2

## --- CheckM lineage workflow ---
## checkm lineage_wf: the standard CheckM pipeline
##   Step 1: tree placement — places each bin into the CheckM reference tree
##   Step 2: lineage-specific markers — selects the appropriate marker gene set
##            for each bin's placement in the tree
##   Step 3: marker gene analysis — finds and quantifies marker genes in each bin
##   Step 4: summary — reports completeness, contamination, strain heterogeneity
##
## positional arguments: input_dir  output_dir
## -t 32   : 32 parallel threads
## -x fa   : file extension for bins (default 'fa')
## --tab_table : output results as a tab-separated table (easier to parse)
## --file checkm_summary.tsv : save summary table to file
checkm lineage_wf \
    4_bins/ \
    5_checkm/ \
    -t 32 \
    -x fa \
    --tab_table \
    --file 5_checkm/checkm_summary.tsv

## --- View results ---
echo "=== CheckM Quality Summary ==="
cat 5_checkm/checkm_summary.tsv

## --- Filter to medium and high quality MAGs ---
## MIMAG standard:
##   High quality: Completeness >= 90%, Contamination <= 5%
##   Medium quality: Completeness >= 50%, Contamination <= 10%
echo ""
echo "=== High Quality MAGs (Completeness >= 90%, Contamination <= 5%) ==="
awk -F'\t' 'NR>1 && $12 >= 90 && $13 <= 5 {print $1, "Completeness:"$12, "Contamination:"$13}' \
    5_checkm/checkm_summary.tsv

echo ""
echo "=== Medium Quality MAGs (Completeness >= 50%, Contamination <= 10%) ==="
awk -F'\t' 'NR>1 && $12 >= 50 && $12 < 90 && $13 <= 10 {print $1, "Completeness:"$12, "Contamination:"$13}' \
    5_checkm/checkm_summary.tsv

## Save lists of HQ and MQ MAG filenames for downstream steps
awk -F'\t' 'NR>1 && $12 >= 90 && $13 <= 5 {print "4_bins/"$1".fa"}' \
    5_checkm/checkm_summary.tsv > 5_checkm/hq_mags.txt

awk -F'\t' 'NR>1 && $12 >= 50 && $13 <= 10 {print "4_bins/"$1".fa"}' \
    5_checkm/checkm_summary.tsv > 5_checkm/hq_mq_mags.txt

echo "HQ MAGs: $(wc -l < 5_checkm/hq_mags.txt)"
echo "HQ+MQ MAGs: $(wc -l < 5_checkm/hq_mq_mags.txt)"

Download:

CheckM metrics decoded

🎯Contamination vs completeness trade-off	A high-quality MAG requires BOTH high completeness AND low contamination. Completeness ≥90% means you recovered most of the genome. Contamination ≤5% means very little foreign DNA is present. You can sometimes improve one metric by splitting a contaminated bin (see CheckM `checkm merge` command), but this risks creating two lower-completeness bins from one contaminated one.
📌Strain heterogeneity	CheckM separates contamination into “strain heterogeneity” (duplicated markers >90% amino acid identity — likely same species, different strain) and true contamination (duplicated markers with <90% identity — different species). High strain heterogeneity can be acceptable for some analyses but reduces confidence in the MAG representing a single genome.

Step 5 — GTDB-Tk Taxonomic Classification

Bash — GTDB-Tk classify workflow to assign taxonomy to MAGs

######################################################################
## STEP 5: Taxonomic classification with GTDB-Tk
## GTDB-Tk assigns taxonomy to MAGs using the Genome Taxonomy Database (GTDB)
## GTDB re-classifies bacteria and archaea based on relative evolutionary
## divergence (phylogenomics), making it more consistent than NCBI taxonomy.
## GTDB-Tk pipeline:
##   1. Identify marker genes (120 bacterial or 53 archaeal single-copy genes)
##   2. Place MAG in the GTDB reference tree using pplacer
##   3. Classify based on tree placement and ANI to reference genomes
######################################################################

module load gtdbtk/2.3.2

## GTDB database must be downloaded separately (~85 GB):
## wget https://data.gtdb.ecogenomics.org/releases/latest/auxillary_files/gtdbtk_data.tar.gz
## tar -xzf gtdbtk_data.tar.gz
## export GTDBTK_DATA_PATH=/path/to/gtdbtk_database/

## --- Run GTDB-Tk classify workflow ---
## gtdbtk classify_wf: full GTDB-Tk pipeline in one command
## --genome_dir : directory containing MAG FASTA files
## --extension fa : file extension for MAG files
## --out_dir   : output directory
## --cpus 32   : parallel threads (pplacer is memory-intensive)
## --mash_db   : path to Mash sketch database (speeds up genome screening)
##   (automatically created by GTDB-Tk if not pre-built)
gtdbtk classify_wf \
    --genome_dir 4_bins/ \
    --extension fa \
    --out_dir 6_gtdbtk/ \
    --cpus 32 \
    --mash_db 6_gtdbtk/mash_db/

## --- View taxonomy results ---
echo "=== GTDB-Tk Bacterial Taxonomy ==="
cat 6_gtdbtk/classify/gtdbtk.bac120.summary.tsv | \
    cut -f1,2,18 | head -30
## Column 1: user_genome (your bin name)
## Column 2: classification (full GTDB taxonomy string)
## Column 18: ANI to closest reference genome

## --- Join CheckM quality + GTDB taxonomy for the final MAG table ---
python3 << 'PYEOF'
import pandas as pd

checkm = pd.read_csv("5_checkm/checkm_summary.tsv", sep="\t")
gtdbtk = pd.read_csv("6_gtdbtk/classify/gtdbtk.bac120.summary.tsv", sep="\t",
                     usecols=["user_genome","classification","closest_genome_reference_radius"])

## Merge on bin name
checkm["bin"] = checkm["Bin Id"]
gtdbtk["bin"] = gtdbtk["user_genome"]
merged = checkm.merge(gtdbtk, on="bin", how="left")

## Extract simplified taxonomy (genus + species from GTDB string)
def parse_gtdb(s):
    if pd.isna(s): return "Unknown"
    parts = {x.split("__")[0]: x.split("__")[1] for x in s.split(";") if "__" in x}
    genus = parts.get("g", "")
    spec  = parts.get("s", "")
    return f"{genus} / {spec}" if spec else genus

merged["taxonomy"] = merged["classification"].apply(parse_gtdb)

## Quality-filtered summary table
summary = merged[["bin","Completeness","Contamination","taxonomy"]].rename(columns={
    "Completeness":"completeness",
    "Contamination":"contamination"
})
summary = summary.sort_values("completeness", ascending=False)
print(summary.to_string(index=False))
summary.to_csv("mag_quality_taxonomy_summary.tsv", sep="\t", index=False)
PYEOF

Download:

Step 6 — dRep Dereplication

Bash — dRep: remove redundant MAGs, keep best representative per species

######################################################################
## STEP 6: Dereplicate MAGs with dRep
## When binning multiple samples or using multiple binners, you often
## recover the same organism multiple times (redundant MAGs).
## dRep clusters MAGs by Average Nucleotide Identity (ANI):
##   Primary cluster: ANI >= 90% (same genus/family level)
##   Secondary cluster: ANI >= 95% (same species level — GTDB species boundary)
## Within each species cluster, dRep selects the BEST MAG
## (highest completeness, lowest contamination).
######################################################################

## Install: conda install -c bioconda drep
module load drep/3.4.0

## Use only high + medium quality MAGs (from CheckM filter in Step 4)
## Low quality MAGs should not be dereplicated — too many artifacts

## --- Run dRep dereplicate ---
## -p 32    : 32 parallel threads
## -comp 50 : minimum completeness for inclusion (matches MIMAG medium quality)
## -con 10  : maximum contamination for inclusion
## -sa 0.95 : secondary cluster ANI threshold (species-level: 95% ANI)
##   This is the GTDB species boundary — genomes within 95% ANI = same species
## --clusterAlg average : linkage algorithm for hierarchical clustering
## -pa 0.9  : primary cluster ANI threshold (genus/family: 90% ANI)
dRep dereplicate \
    7_drep/ \
    -g $(cat 5_checkm/hq_mq_mags.txt | tr '\n' ' ') \
    -p 32 \
    -comp 50 \
    -con 10 \
    -sa 0.95 \
    -pa 0.90 \
    --clusterAlg average

## Output:
##   7_drep/dereplicated_genomes/   ← the best representative MAG per species
##   7_drep/figures/                ← visualisation of clustering
##   7_drep/data_tables/Wdb.csv     ← winner table (which MAG was selected per cluster)
##   7_drep/data_tables/Cdb.csv     ← cluster membership table

echo "=== dRep Summary ==="
echo "Input MAGs: $(cat 5_checkm/hq_mq_mags.txt | wc -l)"
echo "Dereplicated MAGs (unique species): $(ls 7_drep/dereplicated_genomes/*.fa | wc -l)"

Download:

🧬

95% ANI = species boundary: The conventional prokaryotic species threshold is 95–96% Average Nucleotide Identity — genomes above this threshold are considered the same species. GTDB uses 95% ANI as its species-level clustering boundary. When dereplicated, your final MAG set represents each microbial species in your samples exactly once with the best-quality genome.

Step 7 — Prokka Gene Annotation

Bash — Prokka gene prediction and functional annotation per MAG

######################################################################
## STEP 7: Annotate dereplicated MAGs with Prokka
## Prokka: rapid prokaryotic genome annotator
##   1. Prodigal: predict protein-coding genes (ORF prediction)
##   2. BLAST/HMMER: annotate predicted proteins against:
##      - UniProt bacterial reference proteomes
##      - Pfam, Tigrfam protein family databases
##      - rRNA prediction (RNAmmer/Barrnap)
##      - tRNA prediction (Aragorn)
## Output: GFF3, GenBank, protein FASTA, nucleotide FASTA per MAG
######################################################################

module load prokka/1.14.6

## Annotate each dereplicated MAG
for MAG in 7_drep/dereplicated_genomes/*.fa; do

    MAGNAME=$(basename ${MAG} .fa)

    ## Get taxonomy for this MAG from the summary table (for better annotation)
    ## Extract genus from GTDB-Tk classification if available
    GENUS=$(grep ${MAGNAME} mag_quality_taxonomy_summary.tsv | \
            awk -F'\t' '{split($4, a, " / "); print a[1]}')

    ## prokka: annotate prokaryotic genome
    ## --outdir  : output directory for this MAG
    ## --prefix  : output file prefix (used for all output file names)
    ## --genus   : genus name (optional; improves annotation by selecting related DB proteins)
    ## --kingdom : Bacteria or Archaea (affects gene prediction parameters)
    ## --metagenome : use relaxed gene calling for fragmentary metagenomic contigs
    ##   (allows shorter ORFs; important because MAGs may be fragmented)
    ## --cpus 8  : threads for BLAST searches
    ## --norrna  : skip rRNA prediction (faster; rRNA genes are often absent/fragmented in MAGs)
    ## --notrna  : skip tRNA prediction (optional)
    prokka \
        --outdir 8_annotation/${MAGNAME}/ \
        --prefix ${MAGNAME} \
        --kingdom Bacteria \
        --genus "${GENUS:-Bacteria}" \
        --metagenome \
        --cpus 8 \
        --compliant \
        ${MAG} \
        2> 8_annotation/${MAGNAME}_prokka.log

    echo "Annotated: ${MAGNAME}"
    ## Check annotation summary
    grep "CDS" 8_annotation/${MAGNAME}/${MAGNAME}.txt | head -1
done

## --- Summarise annotation across all MAGs ---
echo ""
echo "=== Annotation Summary ==="
for MAGNAME in $(ls 8_annotation/ -d */); do
    MAGNAME=${MAGNAME%/}
    if [ -f 8_annotation/${MAGNAME}/${MAGNAME}.txt ]; then
        CDS=$(grep "^CDS:" 8_annotation/${MAGNAME}/${MAGNAME}.txt | awk '{print $2}')
        HYPO=$(grep "^hypothetical" 8_annotation/${MAGNAME}/${MAGNAME}.txt | awk '{print $2}')
        echo "${MAGNAME}: ${CDS} CDS (${HYPO} hypothetical)"
    fi
done

Download:

Prokka output files decoded

📋.gff	GFF3 format: gene coordinates, product names, and attributes. The most useful output for downstream analyses — use it with featureCounts to quantify gene expression if you later map metatranscriptomic RNA-seq reads to this MAG.
📈.faa	Protein FASTA: the predicted protein sequences for all annotated genes. Use this for functional annotation with eggNOG-mapper (COG/KEGG/GO) or for building custom protein databases for metaproteomics.
🎯Hypothetical proteins %	A well-annotated MAG should have 30–50% of CDS annotated as hypothetical proteins (no known function). If >80% are hypothetical, the bin may represent a very understudied phylum (e.g. CPR/Patescibacteria) or may have assembly errors creating spurious ORFs. Use eggNOG-mapper to get COG functional categories for all genes, including hypothetical proteins.

scRNA-seq Scanpy All Tutorials

Metagenome-Assembled Genomes: MetaBat2 + CheckM

On this page

Overview

Dataset

Directory Structure

Step 1 — QC & MEGAHIT Assembly

Step 2 — Coverage Calculation

Step 3 — MetaBat2 Binning

Step 4 — CheckM Quality Assessment

Step 5 — GTDB-Tk Taxonomic Classification

Step 6 — dRep Dereplication

Step 7 — Prokka Gene Annotation