Free Software for Analyzing Your Own Genome
If you have your own whole-genome sequencing data and want to analyze it, the free stack is straightforward. Use bwa-mem2 or minimap2 for alignment and samtools for BAM handling. Call small variants with DeepVariant or GATK HaplotypeCaller and structural variants with manta plus smoove. Filter with bcftools and annotate with VEP or ANNOVAR. For browsing, use IGV locally. For interpretation of specific variants, use ClinVar, gnomAD, and the ACMG-based aggregators. Every one of those is free, and together they replicate what commercial packages like Geneious wrap in a GUI. The hard part is knowing which files you have, what the numbers in them mean, and where the analysis stops being informative. The software is not the hard part.
This page assumes you have short-read Illumina WGS at roughly 30x coverage, delivered as FASTQ or CRAM. If you only have a consumer genotyping array (a 23andMe or AncestryDNA raw text file), most of this does not apply, and we say why in the last section.
Know which file you have before installing anything
The format determines the analysis, so start by identifying what you received. FASTQ files hold raw reads with per-base quality scores and no genomic position. A 30x human genome arrives as two gzipped FASTQ files (R1 and R2) totaling roughly 100 GB. BAM and CRAM files hold the same reads after alignment to a reference. CRAM is reference-compressed and runs about 3x smaller than BAM, which matters when you are storing several genomes. VCF and its compressed indexed form (.vcf.gz plus .tbi) hold variant calls: positions where your genome differs from the reference.
A common and expensive mistake is analyzing a BAM without knowing its reference build. GRCh37 and GRCh38 coordinates differ by millions of bases in most regions, so annotating GRCh37 calls against a GRCh38 database produces silent nonsense. Check the header:
samtools view -H sample.cram | grep '^@SQ' | head -3
Look at the SN and LN fields. Chromosome 1 with length 249250621 is GRCh37. Length 248956422 is GRCh38. If you have a choice, ask for GRCh38 or the T2T-CHM13 assembly, because the newer references fix thousands of reference errors and add previously missing sequence.
The alignment and calling stack we would use
If you are starting from FASTQ, you need alignment. We use bwa-mem2 for short reads because it is fast, well-validated, and produces output that every downstream caller expects. A minimal pipeline looks like this:
bwa-mem2 mem -t 16 -R '@RG\tID:1\tSM:sample\tPL:ILLUMINA' \
GRCh38.fa R1.fastq.gz R2.fastq.gz \
| samtools sort -@ 4 -o sample.bam -
samtools index sample.bam
The -R read group is not optional. Callers that do base quality recalibration will fail or behave oddly without it. Expect 8 to 20 hours on 16 cores for a 30x genome, and budget 200 GB of scratch space.
Rather than assemble this yourself, consider a packaged pipeline. SpeedSeq was built exactly for this problem. It chains alignment, duplicate marking, and small-variant plus structural-variant calling into a single workflow tuned for one personal genome. It reported whole-genome analysis in around 13 hours on a single server.1 The modern equivalents are nf-core/sarek (Nextflow) and the DRAGEN-compatible open pipelines. We prefer nf-core/sarek because it pins tool versions, resumes after failure, and runs on a laptop-sized dataset or a cloud batch queue with one config change.
For small variants (SNVs and short insertions or deletions), we recommend DeepVariant. It is a convolutional network over pileup images and needs no hand-tuned filtering, which removes the single most error-prone step of the GATK workflow. One command through Docker:
run_deepvariant --model_type=WGS --ref=GRCh38.fa \
--reads=sample.bam --output_vcf=sample.vcf.gz --num_shards=16
GATK HaplotypeCaller remains the reference implementation and is better documented for joint calling across a family. If you are calling one person once, DeepVariant gives fewer false positives for indels with less work.
Structural and copy-number variation need separate tools
Small-variant callers miss deletions, duplications, inversions, and translocations larger than about 50 bases. These are a real fraction of functional variation, so skipping them leaves a gap. We run two callers and intersect them: manta for split-read and discordant-pair evidence, and smoove (a wrapper around lumpy and svtyper) for breakpoint genotyping. Agreement between two orthogonal callers is the cheapest available quality filter.
Copy-number calling from read depth is a different problem and is where low-coverage data can still be useful. WisecondorX detects copy number changes from shallow whole-genome sequencing by normalizing read depth against a reference set of samples, and it works on both single-end and paired-end libraries at coverage far below 30x.2 If you have shallow data from a cheap sequencing run, depth-based CNV calling is the analysis that will still work when variant calling will not.
Long reads change the picture substantially. Oxford Nanopore and PacBio HiFi reads resolve structural variants and repeat expansions that short reads cannot phase or span, and the methods literature now treats long reads as the default for structural and haplotype-resolved work.3 If you are commissioning new sequencing and care about structural variation or phasing, pay for long reads.
Annotation, filtering, and where interpretation stops
A raw 30x genome yields roughly 4 to 5 million variants against GRCh38. Almost all are common and benign. Annotation is how you get from millions of rows to a readable list.
We use Ensembl VEP with the offline cache, because it handles all transcripts for a gene rather than one canonical pick, and because it runs without network calls:
vep --cache --offline --assembly GRCh38 --everything \
--input_file sample.vcf.gz --vcf --output_file annotated.vcf.gz \
--fork 8 --plugin CADD --custom gnomad.genomes.vcf.gz,gnomAD,vcf,exact,0,AF
ANNOVAR is faster and simpler to script, and it is the usual recommendation for a first pass. Its weakness is that gene-based annotation defaults to one transcript, which can hide a consequence that is real in another isoform.
Then filter by population frequency using bcftools. Most pathogenic variants for rare conditions are rare, so a frequency cutoff removes the bulk of the noise:
bcftools view -i 'INFO/gnomAD_AF<0.001 & FILTER="PASS"' annotated.vcf.gz
For the variants that survive, in silico predictors give a first sort. MutationTaster evaluates whether a sequence alteration is likely to cause disease by combining evolutionary conservation, splice site changes, and effects on protein features. It covers synonymous and intronic variants as well as missense.4 Treat all such scores as prioritization. A score is not evidence. A high CADD or MutationTaster score means look here next. It does not mean the variant does anything in you.
This is where free software runs out. Classifying a variant against ACMG criteria requires curated evidence that no local tool contains: segregation data, functional assays, case-control frequencies. Start with ClinVar for existing submissions and check the review status field, because a single-submitter assertion with no assertion criteria is weak. If a variant looks clinically relevant, a clinical geneticist or genetic counselor working from a clinically validated report is the right next step. Research-grade calls are not a diagnosis, and confirmatory clinical testing is standard practice for a reason.
Consumer array files are a different and smaller problem
If your file is a 15 MB text file with roughly 600,000 to 1,000,000 rsIDs, you have array genotypes and not a genome. It covers about 0.02 percent of your bases, chosen to be common variants. You cannot call a novel variant from it, and you cannot detect most structural variation. The right free tool is plink2. It handles the conversion, quality control, and relatedness and ancestry calculations these files support well. Imputation against a reference panel (the Michigan or TOPMed servers) will fill in common variants with good accuracy. It will not recover rare ones, which are the variants most likely to matter functionally.
Questions people also ask
What is the best free software for DNA sequence analysis? There is no single best tool, because the pipeline stages are separable. For personal whole-genome data, the stack we would use is orchestrated by nf-core/sarek. It holds the following tools.
bwa-mem2samtoolsDeepVariantmantabcftools- Ensembl VEP
For short individual sequences such as Sanger reads or a plasmid, use SnapGene Viewer or ApE for viewing and NCBI BLAST for identification.
How do you read DNA sequences? Reading means loading the alignment in a genome browser and inspecting the evidence at a position. Install IGV, load your BAM or CRAM with the matching reference, and navigate to a coordinate. You will see individual reads, mismatches colored by base, and the depth track. Judge a call by depth, allele balance near 0.5 for a heterozygote, and mapping quality.
What is the best tool for aligning DNA sequences? It depends on read length. Use bwa-mem2 for Illumina short reads, minimap2 for Nanopore and PacBio long reads, and MAFFT or MUSCLE for multiple sequence alignment of genes across species. Whole-genome comparisons between species use a different class of aligner again, the approach taken in the mouse-human comparison that identified large conserved syntenic blocks between the two genomes.5
How do I know my variant calls are any good? Run hap.py against the Genome in a Bottle truth sets from NIST. If you have sequenced a reference sample such as NA12878 you can measure precision and recall directly. Otherwise check proxy metrics: mean coverage above 30x, the fraction of the genome at 10x or better above 95 percent, and a heterozygous-to-homozygous variant ratio between about 1.4 and 2.0. Deviations in that ratio usually indicate contamination or a reference mismatch.
Footnotes
-
Colby Chiang, Ryan M Layer, Gregory G Faust, et al. SpeedSeq: ultra-fast personal genome analysis and interpretation. Nature Methods, 2015. https://doi.org/10.1038/nmeth.3505 ↩
-
Lennart Raman, Annelies Dheedene, Matthias De Smet, et al. WisecondorX: improved copy number detection for routine shallow whole-genome sequencing. Nucleic Acids Research, 2018. https://doi.org/10.1093/nar/gky1263 ↩
-
Ying Lu, Mengfei Li, Zhendong Gao, et al. Advances in Whole Genome Sequencing: Methods, Tools, and Applications in Population Genomics. International Journal of Molecular Sciences, 2025. https://doi.org/10.3390/ijms26010372 ↩
-
Jana Marie Schwarz, Christian Rödelsperger, Markus Schuelke, et al. MutationTaster evaluates disease-causing potential of sequence alterations. Nature Methods, 2010. https://doi.org/10.1038/nmeth0810-575 ↩
-
Mouse Genome Sequencing Consortium. Initial sequencing and comparative analysis of the mouse genome. Nature, 2002. https://doi.org/10.1038/nature01262 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.