Manhattan Plot Generator for GWAS Summary Statistics

What does this Manhattan plot generator do?

A Manhattan plot shows every tested variant from a genome-wide association study as a point: genomic position on the x-axis, chromosome by chromosome, and -log10(p) on the y-axis. True associations rise above the background as tall "skyscrapers". This Manhattan plot generator sorts chromosomes naturally (1 to 22, X, Y, MT, with any "chr" prefix removed), converts p-values to -log10 p, and computes cumulative genomic offsets so each chromosome gets a band proportional to its length. It then counts variants above each threshold, identifies lead SNPs, and estimates genomic inflation. Only then does the image model draw the figure, so point heights, threshold lines, and labels come from your data.

Genome-wide significance, suggestive lines, and lead SNPs

• The genome-wide significance threshold of p = 5e-8 is a Bonferroni-style correction for roughly one million independent common-variant tests in European-ancestry genomes. It appears as a dashed line at -log10 p ≈ 7.3.
• The suggestive line at p = 1e-5 (-log10 p = 5) flags loci worth follow-up or replication, but it is not evidence of association on its own.
• Override either line in your instructions, for example "threshold 1e-6", "suggestive 1e-4", "no suggestive line", or "Bonferroni" for 0.05 divided by the number of uploaded variants.
• Lead SNPs are found by distance-based clumping: the most significant variant is kept, every weaker variant within ±500 kb on the same chromosome joins its locus, and the process repeats. Up to 15 leads are labeled. Change the window with "clump window 1 Mb". This is not LD-based clumping; for r²-based loci, run PLINK --clump and upload the result.

Genomic inflation λ and pairing with a QQ plot

• Each p-value is converted to a 1-df chi-square statistic (χ² = z², with z from the normal quantile of p/2). λ is the median χ² divided by 0.4549, the median expected under the null.
• λ close to 1.00 suggests well-calibrated test statistics. Values above about 1.05 can reflect population stratification or cryptic relatedness, but highly polygenic traits in large samples also inflate λ; the LD score regression intercept separates the two.
• A QQ plot compares observed and expected -log10 p. Early deviation from the diagonal points to inflation, while deviation only in the tail usually reflects true signals. Add "QQ plot" to your instructions to place one beside the Manhattan plot with λ printed on it.
• λ is only meaningful when computed from all tested variants. If you upload a filtered subset, compute λ in your analysis pipeline and quote that value.

How to read a GWAS Manhattan plot

• Each column of points is one chromosome, drawn in alternating colors so neighboring chromosomes stay distinct.
• A tower of points reaching above the genome-wide line is a locus: many correlated variants in linkage disequilibrium share the signal.
• An isolated point above the line with no supporting neighbors deserves quality-control checks (low minor allele frequency, poor imputation, or genotyping error).
• Point height reflects statistical evidence, not effect size; report betas or odds ratios separately.

Tips for input files and common mistakes

• PLINK 2 (#CHROM, POS, ID, P), PLINK 1.9 (CHR, SNP, BP, P), REGENIE (CHROM, GENPOS, ID, LOG10P), SAIGE (CHR, POS, MarkerID, p.value), BOLT-LMM (P_BOLT_LMM), and GWAS Catalog (chromosome, base_pair_location, p_value) headers map automatically.
• Columns named LOG10P, NegLog10P, or MLOG10P are treated as already -log10 transformed. Raw p-values must lie between 0 and 1; p = 0 from numeric underflow is clamped to the smallest double and flagged.
• Uploads are capped at 1,200 rows, and extra rows are cut from the end of the file. Filter to variants with p < 1e-3 plus a thinned random sample of the rest, so later chromosomes are not dropped.
• Chromosome widths use GRCh38 lengths by default. Say "GRCh37" for hg19 lengths, or "non-human" or a species name to use observed position spans. PLINK codes 23 to 26 become X, Y, XY, and MT for human data.