Yes, it is possible to perform SNP calling at the assembly level for a bacterial genome, but this approach has significant limitations compared to using raw reads.

Your understanding is correct: analyzing raw reads aligned to a reference genome allows for the incorporation of base quality scores, read depth (coverage), and sequencing platform-specific error profiles, which improve the reliability of variant calls. In contrast, assembled genomes represent a consensus sequence without these supporting metrics, making it difficult to distinguish true SNPs from assembly errors or artifacts.

The general practice in bacterial genomics is to use raw reads for SNP calling whenever possible. This involves aligning reads to a reference using tools like BWA or Bowtie2, followed by variant calling with software such as GATK HaplotypeCaller (for more complex cases), FreeBayes, or samtools/bcftools. For bacterial genomes, specialized pipelines like Snippy or CFSAN SNP Pipeline are common, as they handle haploid genomes and provide core genome SNPs.

If you must use assembled genomes (e.g., due to limited access to raw data), you can align the assembly to a reference and extract differences. Tools for this include:

• MUMmer (specifically, the dnadiff command) for aligning assemblies and reporting SNPs and indels.

  nucmer reference.fasta assembly.fasta
  delta-filter -1 out.delta > out.filtered.delta
  show-snps -ClrT out.filtered.delta > snps.txt
  

• Parsnp for rapid core genome alignment and SNP calling across multiple assemblies.

  parsnp -r reference.fasta -d assemblies_dir -p threads
  

• Alternatively, treat the assembly as long reads: align with minimap2, convert to BAM, and call variants with bcftools (though confidence will be low due to coverage=1).

  minimap2 -ax asm5 reference.fasta assembly.fasta | samtools view -bS - | samtools sort -o aligned.bam -
  samtools index aligned.bam
  bcftools mpileup -f reference.fasta aligned.bam | bcftools call -mv -Ob -o variants.bcf
  

Without read depth, confident variant filtering is challenging; you may need orthogonal validation like PCR. For spurious SNPs, rely on high-quality assemblies and manual inspection.

Kevin