Sequencing data · What you can do here

Call and filter variants from aligned reads

Call, normalise, and filter genetic variants from aligned sequencing reads across explicit genome builds, generating standardised VCF records with calibrated population and functional effect annotations.

Start this in the research agent

What this involves

Variant calling from aligned reads transforms BAM or CRAM alignments into discrete records of genomic variation. The process requires rigorous tracking of genome assemblies and coordinate conventions. Discrepancies between GRCh37, GRCh38, and T2T-CHM13v2.0, or confusion between 0-based half-open indexing (BAM, BED) and 1-based closed coordinates (VCF, GFF3), cause silent positional errors. Consistent chromosome naming, such as reconciling chr1 with 1, is essential before running overlap operations.

Post-calling processing demands systematic indel normalisation. Because VCF conventions left-align indels whilst HGVS standards 3'-shift within repeat sequences, raw call sets frequently encode identical biological variants under differing coordinate representations. Splitting multi-allelic records and left-aligning indels against the reference sequence ensures call sets can be reliably queried against reference databases.

Filtering variant sets relies on calibrated population frequencies and orthogonal functional predictors. Evaluating the ancestry-group popmax allele frequency in gnomAD rather than global averages isolates rare alleles relevant to Mendelian analysis. In silico impact models like AlphaMissense, REVEL, CADD, and SpliceAI must be applied as single calibrated lines of evidence within standard classification frameworks, supported by verified clinical assertions from ClinVar.

How it goes

1Alignment inspection and header validationInspect and index alignment files with samtools 1.20, confirming reference contig names and assembly coordinates to avoid silent zero-overlap errors during downstream queries.
2Variant callingIdentify single-nucleotide variants and indels across targeted regions using bcftools 1.20, or execute end-to-end germline and somatic workflows via nf-core/sarek or nf-core/raredisease.
3Indel normalisation and multi-allelic splittingNormalise variant representations with bcftools norm against the target reference FASTA, splitting multi-allelic calls and left-aligning indels to match standard VCF conventions.
4Population allele frequency filteringCross-reference variant alleles against gnomAD through GraphQL queries to evaluate ancestry-group popmax allele frequencies and filter out common non-pathogenic variation.
5Effect prediction and clinical annotationAnnotate variant consequences using Ensembl VEP, evaluating calibrated scores from AlphaMissense, REVEL, CADD, or SpliceAI alongside ClinVar submissions retrieved via NCBI E-utilities.

Start it here

Ask in plain language. The agent works on a Linux machine of its own — it installs and runs what the job needs, keeps every file, and reads the results back with the numbers explained. Nothing to set up, and a long run continues after you close the page.

“Sort and index my BAM file with samtools and report the mapping statistics.”Runs samtools “Filter this VCF to high-quality SNPs with bcftools and count them per chromosome.”Runs bcftools “Map my reads to this reference with bwa mem and report the alignment rate.”Runs BWA “Align these paired-end reads to the reference with Bowtie 2.”Runs Bowtie 2

Open the research agent

What to watch for

A coordinate without an explicitly defined genome build is unusable because loci can shift by megabases between GRCh37, GRCh38, and T2T-CHM13v2.0.
Mixing 0-based half-open coordinates from BAM or BED files with 1-based fully closed VCF coordinates introduces off-by-one errors.
Chromosome naming mismatches between chr1 and 1 or chrM and MT cause joins to fail silently with zero overlaps and no error messages.
Failing to split multi-allelic sites and left-align indels prevents accurate matching, as VCF left-aligns while HGVS 3'-shifts in repetitive regions.
Averaging multiple correlated missense predictors produces redundant metrics rather than independent evidence, counting as only one supporting line (PP3/BP4) in ACMG guidelines.
Filtering on global allele frequency instead of the ancestry-group popmax allele frequency in gnomAD risks overlooking common stratified alleles.

What it runs

Programs

Databases it can query

nf-core pipelines

Questions

Do I need to install anything to call and filter variants from aligned reads?

No. The research agent does this work on a Linux machine of its own — it runs samtools, bcftools, BWA, Bowtie 2 there, keeps the files, and reads the results back to you. Long runs continue if you close the page.

Why is indel normalisation necessary after calling variants?

In repetitive tracts, VCF convention left-aligns insertions and deletions whereas HGVS nomenclature shifts them 3'. Splitting multi-allelic records and left-aligning with bcftools norm ensures identical physical mutations share a uniform representation for reliable comparison.

How should computational variant effect scores be weighed?

Select a single calibrated predictor per functional class—such as AlphaMissense or REVEL for missense changes and SpliceAI for splicing—rather than averaging correlated tools. Under ACMG guidelines, these tools provide only a single supporting line of evidence (PP3/BP4).

How do coordinate systems differ between BAM, BED, and VCF files?

BAM files and BED formats use 0-based, half-open intervals where the start coordinate is offset by one, whereas VCF and GFF3 files use 1-based, fully closed intervals. A SNV at 1-based position 100 must be represented as start 99 and end 100 in BED format.

What precautions are required when lifting over coordinates between genome builds?

Liftover using chain files carries an expected 0.1% to 1% failure rate that indicates unstable or missing regions rather than ignorable records. Following liftover, the reference allele must be re-derived from the destination FASTA because strand orientation can flip.

Written from the method the research agent follows for this work and the documentation of the tools listed above. Statements citing a figure that does not appear there are removed before publication. Corollary is a non-commercial research demonstration; results are not for clinical use.