SNPstats jamovi module

The SNPstats jamovi module provides a complete pipeline for single-SNP and multi-SNP genetic association analyses.

jamovi is a free and open statistical spreadsheet, built on R, that runs on Windows, macOS and Linux (and in the browser, as jamovi cloud).

This module extends jamovi’s functionality to bring the analyses of the SNPstats web tool into jamovi, so the data never leaves your computer, datasets are not limited by the web upload, and results can be combined with the rest of jamovi’s analyses and with jamovi’s data editing, filtering and recoding.

How to install

  1. Install jamovi — download it from www.jamovi.org/download.html and install it as any other application.
  2. Install the module in jamovi — open jamovi and click the + (Modules) icon at the top right of the window, choose Jamovi library, write SNPstats in the search, and click INSTALL.
  3. Check the installation — an SNPstats menu appears in the Analyses ribbon, with the SNP Analysis and Polygenic Score (PGS) entries described below.

Open source

The code for this module can be found in GitHub. Report there any bug, inconsistency, or improvement suggestions.

Overview

SNPstats jamovi is organised into two modules accessible from the menu:

Data format

Each row in the dataset represents one individual. SNP genotypes must be stored as character columns using diploid notation — two alleles separated by a delimiter, or concatenated as two characters. The module accepts the following formats:

Format Separator Example values
A/B slash / T/T, T/C, C/C
A|B pipe | T|T, T|C, C|C
A>B greater-than > T>T, T>C, C>C
AB none (2 chars) TT, TC, CC

Any allele names are accepted (single nucleotides, insertion/deletion codes, etc.), provided that exactly two distinct alleles appear across the column. Columns with three or more alleles, or inconsistent genotype combinations (e.g. A/A, A/C, B/C) are flagged and excluded with an informative message.

Allele order is initially defined by the frequency, sorting the most frequent allele first so it becomes the reference for analysis of association. This can be changed in Jamovi Data panel.

The response variable should be a binary (case/control coded 0/1 or as a two-level factor), categorical (multi-level factor) or continuous column. Covariates can be numeric or categorical.

SNP Analysis module

The SNP Analysis module (Figure 1) is the starting point for any analysis. Assign the response variable, one or more SNPs, and any covariates in the shared variable assignment panel at the top.

There are two global options, that affect multiple analyses:

Figure 1 – SNP Analysis options panel Figure 1. SNP Analysis options panel showing shared variable assignment, response type, and the three analysis tabs.

Descriptive tab

The Descriptive tab provides genotype and allele frequency summaries, Hardy–Weinberg equilibrium tests, and covariate descriptions.

Covariate descriptive

Enable Descriptive statistics for covariates to obtain a covariate summary table Figure 2. Each variable is summarised as follows:

When Stratify by response groups is active and the response is binary or categorical, the table gains separate columns for each group plus a p-value column: χ² test for categorical variables and independent-samples t-test or ANOVA for continuous variables.

Figure 2 – Covariate Descriptives output Figure 2. Covariate summary stratified by case/control status (STATUS). The p-value column uses χ² for SEX and BMI (categorical) and a t-test for AGE (continuous).

In the example dataset (n = 706; 329 controls, 377 cases), none of the covariates differ significantly between groups: SEX (p = 0.0969), AGE (p = 0.252), BMI (p = 0.122).

SNP descriptive

Enable SNP summary table to obtain a compact overview of all SNPs in a single table Figure 3. For each SNP and group the table reports:

Figure 3 – SNP Summary Table Figure 3. SNP summary table for SNP1 (C/G) and SNP2 (C/T), overall and stratified by case/control status.

In the example, SNP1 (C/G) has MAF = 0.361 and is in HWE overall (p = 0.247). SNP2 (C/T) has MAF = 0.609 and shows a significant deviation from HWE in the full sample (p = 0.002) and among cases (p = 0.012), warranting further quality-control investigation.

Detailed outputs can be toggled independently:

Checkbox Output produced
Allele frequencies Per-SNP allele counts and proportions, stratified when applicable
Genotype frequencies Per-SNP genotype counts and proportions, stratified when applicable
Hardy–Weinberg equilibrium Exact HWE test p-value, overall and per stratum
Show missing values Adds a Missing row to frequency tables
Show SNP missingness plot displays a bar chart of missing genotype rates per SNP; only SNPs exceeding the Show SNPs with missingness above (%) threshold are plotted

Allele and genotype frequencies

Enable Allele frequencies and Genotype frequencies for per-SNP detailed tables (Figure 4). Counts and proportions are computed over typed individuals only; missing genotypes are excluded from denominators but reported separately when Show missing values is checked.

Allele frequencies are derived by counting each allele separately across both chromosomes:

\[\text{Allele freq}(A) = \frac{2 \times N_{AA} + N_{AB}}{2 \times N_{\text{typed}}}\]

Figure 4 – Allele and Genotype Frequencies for SNP1 Figure 4. Allele frequencies, genotype frequencies, and HWE test for SNP1 (C/G). The C allele is more common (63.9%). Genotype distributions are similar between controls and cases.

Hardy–Weinberg equilibrium

Enable Hardy–Weinberg equilibrium to obtain an exact test p-value for each SNP. Under HWE, allele frequencies within a large, randomly mating population remain constant across generations, and genotype proportions follow the expected binomial distribution. A significant departure may indicate genotyping error, population stratification, or selection at or near the locus.

When Stratify by response groups is active, HWE is tested separately within controls and cases (Figure 4, bottom panel). Deviation from HWE in controls only is a particular concern for genotyping quality, whereas deviation only in cases could reflect genuine selection.

In the example, SNP1 does not depart from HWE in any group. SNP2 shows significant deviation overall and in controls, which may warrant exclusion or additional quality checks before proceeding to association analysis. HWE in cases only may be indicative of an association with the disease.

SNP missingness plot

Enable Show SNP missingness plot to display a bar chart of missing genotype rates per SNP; only SNPs exceeding the Show SNPs with missingness above (%) threshold are plotted.

Figure 5 – SNP Missingness Plot Figure 5. SNP missingness plot. Only SNPs exceeding the missingness threshold are plotted. The axis is in % scale.

Association tab

The Association tab tests each SNP individually for association with the response and, optionally, for interaction with a covariate.

Figure 6 – Association tab options Figure 6. Association tab options. Genetic models, confidence interval width, AIC/BIC display, and interaction sub-analyses are all configurable.

When covariates are present, every model is adjusted for them. The number of observations excluded due to missing covariate values is reported in a table note.

Association with outcome

Enable SNP-response association to fit a regression model — logistic for binary, linear for continuous — for each SNP under each selected genetic model (Figure 7). Let A be the reference allele (the allele in the most frequent homozygote) and B the variant allele. The five available models are:

Model Comparison encoded Interpretation
Codominant AA = 0, AB = 1, BB = 2 (two dummy variables) Tests each heterozygote and homozygote separately vs. AA
Dominant AA = 0, AB + BB = 1 Any copy of B increases risk equally
Recessive AA + AB = 0, BB = 1 Two copies of B required for effect
Overdominant AA + BB = 0, AB = 1 Heterozygote advantage/disadvantage
Additive 0, 1, 2 copies of B Each additional B allele multiplies OR by a constant factor (log-additive on the logit scale for a binary response fitted by logistic regression)

For each model the table reports: genotype group counts by case/control status, OR (or β for linear), lower and upper confidence interval bounds, a p-value, AIC, and BIC. Each model’s p-value is the likelihood-ratio test (binary/categorical) or F-test (quantitative) for overall association — comparing that genetic model against the covariate-only null — shown once on the model’s reference row (two degrees of freedom for the codominant model, one for the others); the individual genotype rows do not carry a separate p-value. This matches the p-values reported by the SNPstats web tool. AIC and BIC facilitate model comparison across genetic models — a lower value indicates a better-fitting, more parsimonious model.

Figure 7 – SNP Association Results for SNP1 Figure 7. SNP Association Results for SNP1 adjusted for SEX, AGE, and BMI. Each model reports one p-value (the likelihood-ratio test) on its reference row; the individual genotype rows carry only the OR and CI. No model shows a significant association (LRT p = 0.236 for Codominant). The Overdominant model has the lowest AIC (948.08), though differences are small.

In the example, SNP1 shows no significant association with STATUS under any model after adjustment for SEX, AGE, and BMI (all p > 0.09). The additive (log-additive for this logistic model) OR per G allele is 0.956 (95% CI: 0.768–1.190, p = 0.689), consistent with a null effect.

SNP × covariate interaction

Enable SNP × covariate interaction to test whether the SNP effect on outcome differs across levels of the first covariate listed (Figure 8). The interaction model is:

\[\text{logit}(p) = \beta_0 + \beta_1{\text{SNP}} + \beta_2{\text{Z}} + \beta_{\text{int}}{SNP \times Z}\]

The interaction table lists OR and 95% CI for SNP main effects, the covariate main effect, and each SNP × covariate product term $\beta_{\text{int}}$, that is a vector ans size depends on the genetic model and covariate categories. The Interaction p-value (LRT) compares the full interaction model against the additive main-effects model and is reported in a table note. The Interaction parameterisation dropdown offers three options:

Detailed tables for these models are shown below.

The Show adjustment covariate parameters checkbox includes the remaining covariates’ coefficients in the table.

Figure 8 – Interaction Results (Codominant model, SNP1 × SEX) Figure 8. SNP1 × SEX interaction under the codominant model. The interaction LRT p-value is 0.915, indicating no evidence that the SNP1 effect differs between females and males.

Stratified analysis by covariate

Enable Stratified analysis by covariate to parametreize the association model within each level of the interaction covariate (Figure 9). Each stratum table uses the same reference genotype as the pooled analysis, facilitating direct comparison of effect estimates across strata.

Figure 9 – Stratified Analysis by SEX Figure 9. SNP1 association with STATUS stratified by SEX. Effect estimates are consistent between females and males, corroborating the non-significant interaction (p = 0.915).

Stratified analysis by genotype

Enable Stratified analysis by genotype to flip the stratification: the covariate effect on outcome is estimated separately within each genotype group (Figure 10). The reference level of the covariate is held constant across strata.

Figure 10 – Stratified Analysis by Genotype Figure 10. SEX effect on STATUS within each SNP1 genotype group. Males show a consistently higher OR than females across all genotypes, with no significant variation by genotype.

Cross-classification table

Enable Show cross-classification table to display a full factorial breakdown of case/control counts and ORs for every combination of genotype and covariate level (Figure 11). The reference cell is the combination of the reference genotype and the reference covariate level.

Figure 11 – Cross-Classification: SNP1 × SEX Figure 11. Cross-classification of SNP1 genotype × SEX with ORs relative to the Female / C/C reference cell. All ORs are consistent with no interaction (LRT p = 0.915).

LD and Haplotype tab

When two or more SNPs are assigned, the LD and Haplotype tab becomes available. It combines linkage disequilibrium statistics with haplotype frequency estimation and association testing.

Figure 12 – LD and Haplotype tab options Figure 12. LD and Haplotype tab with five SNPs, three covariates, and all sub-analyses enabled.

Linkage disequilibrium

Linkage disequilibrium (LD) is the non-random association of alleles at different loci on the same chromosome. It arises because chromosomes are inherited as blocks, with recombination between neighbouring loci being rare. SNPs in high LD with a causal variant serve as proxies for it, making LD analysis central to fine-mapping and tag-SNP selection.

Enable Pairwise LD table to compute three statistics for every pair of SNPs (Figure 13):

Statistic Range Interpretation
D (−∞, +∞) Raw covariance of allele frequencies; magnitude depends on allele frequencies
D′ [0, 1] Scaled D; D′ = 1 means no recombination has been observed between the two alleles
r² [0, 1] Squared correlation; r² = 1 means the two SNPs are perfectly interchangeable as tag SNPs

A p-value testing departure from linkage equilibrium (D = 0) is also reported.

Enable LD matrix to display the pairwise r² values (or whichever metric is selected in the Metric dropdown) in a square matrix, with p-values in the lower triangle and SNP names on the diagonal (Figure 13, lower panel).

Enable LD heatmap for a colour-coded visualisation of the matrix (Figure 14). Cells in the upper triangle show r² values and are shaded from white (r² = 0) to dark red (r² = 1); cells in the lower triangle show p-values.

Figure 13 – Pairwise LD Table and LD Matrix Figure 13. Pairwise LD results for five SNPs. SNP1 and SNP2 show high LD (r² = 0.847, D′ = 0.981). SNP3 and SNP4 show moderate LD (r² = 0.537). All pairs are statistically significant (all p < 0.001).

Figure 14 – LD Heatmap Figure 14. LD heatmap for five SNPs. The SNP1–SNP2 block (top-left) is clearly visible. SNP3–SNP4 form a secondary block. SNP5 shows moderate LD with SNP1 and SNP2 but low LD with SNP3 and SNP4.

Haplotype frequencies

A haplotype is the combination of alleles across multiple loci on a single chromosome. Because standard genotyping does not resolve which alleles reside on the same chromosome (phase is unknown for doubly heterozygous individuals), haplotypes must be inferred statistically using the Expectation–Maximisation (EM) algorithm, which iterates between:

  1. E-step: compute the posterior probability of each possible haplotype pair for each individual, given current frequency estimates.
  2. M-step: update haplotype frequency estimates as the weighted average of haplotype assignments across individuals.

Enable Haplotype frequency estimation to run the EM algorithm and obtain a haplotype frequency table (Figure 15). The Min freq for rare haplotypes field (default 0.01) pools all haplotypes with estimated frequency below the threshold into a single Rare category. When Stratify by response groups is active (the shared option in the SNP Analysis panel), frequencies are shown separately for controls and cases.

Figure 15 – Haplotype Frequencies Figure 15. Haplotype frequency table for five SNPs. The most common haplotype is C-T-C-C-G (freq = 0.594), followed by G-C-C-C-T (0.225). Haplotype C-C-A-G-G is enriched in cases (0.035) relative to controls (0.014).

In the five-SNP example, six distinct haplotypes exceed the 0.01 threshold plus one Rare pooled category. The two dominant haplotypes together account for about 82% of chromosomes.

Haplotype association

Enable Haplotype-response association to test each haplotype for association with the outcome using haplo.glm from the haplo.stats package (Figure 16). This method fits a regression model weighted by the posterior probability of each haplotype assignment for every individual, correctly propagating phase uncertainty into the parameter estimates.

The most frequent haplotype is automatically selected as the reference category. The table reports OR (binary outcome) or β (continuous outcome) with confidence interval and p-value for each common haplotype, plus the pooled Rare term. When covariates are included, the model adjusts for them. regarding sample size used, haplotypes estimated with the EM algorithm allow for missing values in some SNPs. Only cases with all SNPs missing are excluded.

Figure 16 – Haplotype Association Figure 16. Haplotype association with STATUS, adjusted for SEX, AGE, and BMI. Haplotype C-C-A-G-G shows a significant association (OR = 2.547, 95% CI: 1.132–5.73, p = 0.024) relative to the reference C-T-C-C-G. No other haplotype reaches significance.

Haplotype × covariate interaction

Enable Haplotype × covariate interaction to test whether haplotype effects differ across levels of the first covariate (Figure 17). The interaction model is:

\[\text{logit}(p) = \beta_0 + \sum_h \beta_h H_h + \beta_Z Z + \sum_h \beta_{h \times Z}(H_h \times Z)\]

The table lists main haplotype effects, the covariate main effect, and each haplotype × covariate product term with OR, 95% CI, and p-value. An overall likelihood ratio test for interaction is reported in a table note, computed as the deviance difference between the full interaction model and the main-effects-only model.

Figure 17 – Haplotype × Covariate Interaction Figure 17. Haplotype × SEX interaction. No haplotype × SEX interaction term is individually significant, and the overall LRT for interaction is p = 0.945, indicating that the haplotype effects do not differ by sex.

Polygenic Score (PGS) module

A polygenic score (also called a polygenic risk score, PRS) aggregates the effects of many genetic variants across the genome into a single numerical index for each individual. Rather than testing one SNP at a time, a PGS weights each variant’s dosage (0, 1, or 2 copies of the effect allele) by its published effect size and sums the products. Higher scores indicate a higher genetic burden of the trait in question relative to the study sample.

The Polygenic Score module (Figure 18) can operate in two modes: a fully weighted mode that reads effect sizes from an external weights file (e.g. from the PGS Catalog), and an unweighted mode that assigns unit weights to all selected SNPs. Both modes can be run simultaneously for comparison.

Figure 18 – PGS module options panel Figure 18. PGS module options panel showing variable assignment, weighting options, QC filters, scoring settings, and analysis sub-sections.

The example file CRCgenet-SNPs.tsv will be analyzed, that contains data on 65 SNPs associated with colorectal cancer (CRC). The data were measured in 2838 individuals. This file has been generated permuting randomly real data, to preserve phenotype and sex, but other covariates have been generated randomly. The phenotype corresponds to high-risk adenoma or cancer detected in a study of CRC screening.

Variable assignment

Assign one or more SNP columns (genotype data) to the SNP columns box. The same genotype formats accepted by all other modules are supported (slash-, pipe-, or concatenated two-character notation). Numeric dosage columns (values 0, 1, 2) are also accepted, though allele QC cannot be performed on them.

Optionally assign a response variable (binary or continuous) and one or more covariates. These are used only for the analysis sub-sections (association, percentile regression, interaction); the score itself is computed from the SNP columns alone.

SNP coverage summary

Enable SNP coverage summary to display a table that characterises the SNPs and shows some QC checks. This table is updated if a weight file is uploaded, including the match between the weights file and the dataset. When a weights file is loaded, the row Risk allele ≠ minor allele counts how many of the scored SNPs have their published risk (effect) allele equal to the major (more common) allele in your data — i.e. SNPs for which the frequency-based default orientation (used when no file is loaded) would have pointed the opposite way. A high count is expected and harmless (risk alleles are often common); it simply quantifies how much the weights file re-orients relative to the no-file default.

SNP QC filters

Two optional filters can exclude SNPs from the score before it is computed:

Both filters are applied after per-SNP statistics have been computed, so QC metrics are always shown regardless of whether the SNP is subsequently excluded. To remove SNPs that do not pass QC from the table, use Show valid SNPs only.

Weighting

Weights file

Click the 📁 button next to the Weights file field to choose a file in PGS Catalog scoring file format. The file may be plain text (.csv, .tsv, .gz) with tab, semicolon or comma separators.

The field itself is read-only and shows only the chosen file’s name: the module stores the file’s contents in the analysis, never its location. This is why it works identically in jamovi cloud, where the R engine runs on a different machine than the browser and a local path would mean nothing — and it is also what keeps a saved .omv safe to share, since re-opening it cannot make jamovi read a file from the new reader’s own computer.

From R, use the pgs_weights() helper, which reads the file in your session and returns the two arguments that carry it:

w <- pgs_weights("CRCgenet-PGS.txt")
do.call(snpPGS, c(list(data = mydata, snpCols = c("rs1", "rs2")), w))

Header comment lines beginning with # are parsed for catalogue metadata (PGS ID, score name, trait, weight type, genome build, variant count); these appear in the SNP Coverage Summary table.

Required columns are rsid (or variant_id), effect_allele, other_allele, and effect_weight. Chromosomal position (chr_name, chr_position) and any additional columns present in the file are retained and displayed in the SNP Weights table. The module matches file entries to dataset columns by the SNP column name; names that appear in the file but not in the dataset are flagged as unmatched.

For our example data, file CRCgenet-PGS.txt has been generated with the weights and effect alleles for the 65 SNPs. The file in fact includes an additional SNP not present in the dataset.

SNP weighting mode

The SNP weighting mode dropdown controls which score variants are computed:

When no weights file is provided, the module automatically falls back to unweighted scoring regardless of the dropdown selection.

Important — the weights file also orders the alleles. Beyond supplying effect sizes, the weights file defines each SNP’s effect (risk) allele, which is the allele the dosage counts. Without a file the module cannot know the true risk allele, so it assigns one from the data using the same rule as the SNP Analysis module: the minor (less frequent) allele is taken as the effect allele, unless you have reordered the genotype levels in the Jamovi Data panel, in which case that order is respected (the first genotype level’s homozygote allele becomes the reference and the other becomes the effect allele). This makes the two modules consistent, but the direction is only a convention, not a validated risk orientation — for any given SNP the published risk allele may be the major allele, so a frequency-assigned effect allele can point the opposite way. The unweighted score then sums risk-increasing and risk-decreasing directions together and any real association tends to wash out, which is why an unweighted result can jump from non-significant to significant once a weights file is loaded — the file re-orients every SNP to its true risk allele. Always load a weights file (even one with all effect weights set to 1 if you only want an allele count) so the alleles are oriented to the published risk allele; a no-file score is not a genuine polygenic risk score and the result tables carry a warning to that effect.

Scale file weights to unit mean

Enable Scale weights to unit mean to L1-normalise the weight vector so that its mean absolute value equals 1 before scoring. This places the weighted and unweighted scores on a comparable numeric scale (both approximate a risk-allele count), making the “Both” mode more informative for direct comparison.

SNP Weights Used for Scoring (Figure 19) shows one row per SNP from the weights file. Columns include chromosomal position (if available in the file), effect and other alleles, effect weight, match status, and allele QC outcome. Per-SNP statistics computed from the dataset — missingness rate (N and %), effect allele frequency, and HWE p-value — are shown alongside any additional columns from the weights file. The Allele QC column uses a traffic-light coding:

Icon Meaning
✅ SNP matched and alleles verified
⚠ SNP kept with an action taken (e.g. strand flip, numeric dosage)
❌ SNP excluded (allele mismatch, multiallelic, all missing, etc.)

Enable Show valid SNPs only to hide rows for SNPs that failed QC, showing only those that contributed to the score.

Figure 19 – SNP Weights Table Figure 19. Fraction of SNP Weights Used for Scoring table. Each row corresponds to one SNP from the weights file. The Allele QC column summarises the per-SNP quality control outcome.

SNP Coverage Summary (Figure 20) condenses the match statistics into a key–value table. It reports: metadata from the file header (PGS ID, trait, genome build); the number of SNPs in the file, in the dataset, and successfully matched; counts of ambiguous (A/T or C/G) SNPs, strand-flipped SNPs, allele mismatches, and null-allele genotypes fixed; SNPs excluded by QC filters; and final sample-size information including the number of complete cases (individuals with observed genotypes for all scoring SNPs).

Figure 20 – SNP Coverage Summary Figure 20. SNP Coverage Summary table for an example PGS Catalog score. Metadata rows are shown only when the weights file includes a header.

Scoring options

Missing genotype handling

The Missing genotype handling dropdown specifies how missing genotypes are treated when computing the weighted sum:

Strategy Description
Use observed SNPs only (SNP-wise) Each individual is scored using only their observed SNPs; score is automatically adjusted for coverage (default)
Mean imputation Missing dosage is replaced by the SNP’s observed mean dosage across all individuals
Treat as 0 Missing dosage is set to 0 (equivalent to assuming the reference homozygote)
Exclude individuals Any individual with at least one missing genotype across the scoring SNPs is excluded from all analyses

Correct for missing genotypes

Enable Correct for missing genotypes to divide each individual’s raw weighted sum by their per-individual theoretical maximum — the sum of effect weights for SNPs with observed genotypes only. This correction removes the downward bias introduced when some SNPs are missing: an individual typed at 90% of SNPs should not receive a lower score simply because 10% are unobserved. The correction is strongly recommended when the proportion or percent rescaling methods are used, as those methods are only valid when the denominator reflects the individual’s actual coverage.

Score rescaling

The Score rescaling dropdown applies a transformation to the (optionally corrected) score:

Method Formula Typical use
None Raw weighted sum When downstream software expects raw allele-count units
Proportion Score / theoretical max → [0, 1] Comparable across scores with different numbers of SNPs
Percent Score / theoretical max × 100 → [0–100] Human-readable proportion
Multiply Score × user factor Arbitrary scaling (e.g. per 10 risk alleles)
Per N alleles Score / N Expresses the score as a per-allele average

For Multiply and Per N alleles, the Scale factor / N alleles field sets the multiplier or the N.

Standardize score (SD = 1)

Enable Standardize score to divide the final score by its sample standard deviation. The resulting score has SD 1 (the mean is left unchanged, not centred to 0), placing weighted and unweighted scores on a common standard-deviation scale and enabling OR and β estimates from the association table to be interpreted as “per-SD” effect sizes.

PGS Summary Statistics

The PGS Summary Statistics table (Figure 21) is shown by default whenever at least one SNP is successfully scored; it can be hidden with the Show PGS summary statistics switch. For each score type (Weighted, Unweighted, or both), it reports descriptive statistics of the score distribution: N, mean, SD, 95% confidence interval of the mean, minimum, maximum, and skewness. When Stratify summary by response is enabled (the default) and a binary or categorical response variable is assigned, separate rows are shown for each outcome group plus an overall row; disable it, or use a continuous response, for a single overall row per score type.

Figure 21 – PGS Summary Statistics Figure 21. PGS Summary Statistics table showing descriptive statistics of the score distribution overall and by case/control group.

Save PGS scores to data

Open the Save section at the bottom of the options panel to write the computed scores back to the jamovi dataset as new continuous columns — one per active weighting mode, named PGS_Weighted and/or PGS_Unweighted. Once saved, these columns are available to every other jamovi module for additional analyses — for example, as a predictor in a linear model or as an outcome in a mediation analysis. Scores are aligned to the correct original rows even when the Exclude individuals missing-genotype strategy is active and some rows were dropped during scoring.

Analysis sub-sections

The three analysis sub-sections are enabled independently and require a response variable to be assigned. All analyses operate on complete cases with respect to the PGS score, the response, and any covariates.

PGS – response association

Enable PGS – response association test to regress the response on the PGS score (Figure 22). The set of tests run depends on the response type detected automatically from the data:

Binary response (two levels): logistic regression reporting the OR per unit PGS, Welch two-sample t-test of PGS means between cases and controls (reporting the mean difference), and Mann–Whitney U test (reporting the Hodges–Lehmann estimator of the location shift).

Continuous response: linear regression reporting the slope (β) and standard error, Pearson correlation (r), and Spearman rank correlation (ρ). For Spearman correlation, 95% confidence intervals are computed from the Fisher z-transformation.

Polytomous response (more than two levels): multinomial logistic regression (via nnet::multinom) with one OR row per non-reference level relative to the lowest level, plus an overall likelihood ratio test across all levels; one-way ANOVA (F statistic); and Kruskal–Wallis test (χ²).

When covariates are assigned, the regression models adjust for them; parametric tests are unadjusted. Table notes identify the response coding and the covariates included.

Figure 22 – PGS–Response Association Figure 22. PGS–response association table. For a binary response the table shows the logistic regression OR, the t-test mean difference, and the Mann–Whitney location-shift estimate, each with 95% CI and p-value.

PGS percentile category analysis

Enable Show percentile ranks to divide the score distribution into percentile-defined categories and analyse outcomes across them (Figure 23).

The Percentile thresholds field accepts a comma-separated list of percentile cut-points (e.g. 20,40,60,80,90,95), which define N+1 categories labelled <P20, P20–P40, …, >P95. Any values outside the 0–100 range or duplicate values are silently dropped; if the field is left blank, the default cut-points 20, 40, 60, 80, 90, and 95 are used.

The Reference category dropdown selects which category serves as the baseline in regression models: lowest, highest, or middle (the category containing the sample median; default).

Two tables are produced:

PGS Percentile Category Counts (Figure 23, left) shows, for each category: the score range observed within it, overall N (%), and — when a binary or polytomous response is selected — a column for each outcome level.

PGS Percentile Category Analysis (Figure 23, right) runs a regression of the response on the category factor. For a binary response, logistic regression yields ORs with 95% CI and p-value for each non-reference category. For a continuous response, linear regression yields mean differences (β). The reference category is marked with a ◆ symbol and its estimate is fixed to 1 (binary) or 0 (continuous). Covariates are included in the regression model when assigned.

Figure 23 – PGS Percentile Category Tables Figure 23. Percentile category counts (left) and category analysis (right). Each row corresponds to one score percentile band. The reference category (◆) has OR = 1 by definition.

PGS × covariate interaction

Enable PGS × covariate interaction to test whether the PGS effect on the response differs across levels of the first covariate in the covariate list (Figure 24). At least one covariate and a response variable must be assigned for this option to be enabled.

The interaction model is:

\[g(\mu) = \beta_0 + \beta_{\text{PGS}} \cdot \text{PGS} + \beta_Z \cdot Z + \beta_{\text{PGS} \times Z} \cdot (\text{PGS} \times Z)\]

where $g$ is the logit link for binary outcomes and the identity link for continuous outcomes. Additional covariates beyond the first are included as additive adjustment terms.

The table reports estimates (OR for binary, β for continuous) with 95% CI and p-value for: the PGS main effect, the first covariate’s main effect (one row per non-reference level if categorical), and each PGS × covariate product term. A likelihood ratio test comparing the full interaction model against the main-effects-only model is reported in the final row. A table note lists any additional adjustment covariates.

Figure 24 – PGS × Covariate Interaction Figure 24. PGS × covariate interaction table. The table reports the PGS and covariate main effects and each interaction product term from the same regression model, together with the overall LRT p-value for the interaction.

Plots

Enable individual plots in the Plots section to visualise score distributions and model performance. For a binary or categorical response, the Case / event level dropdown selects which response level is treated as the case/event — the positive class for the ROC and calibration curves and the reference group for the distribution colours; leaving it empty uses the last observed level.

Score distribution plot

Enable Show score distribution plot to visualise the PGS distribution (Figure 25). The Plot type dropdown offers three options:

When a binary response variable is assigned, separate distributions are drawn for each outcome group using distinct colours, a legend identifies the groups, and a Mann–Whitney p-value (two-sided) comparing the two distributions is annotated above the plot.

When a continuous response is assigned, an additional PGS vs Response scatter plot is produced showing the raw data points with an ordinary least-squares regression line, R², and p-value annotated.

Dashed vertical lines mark the group means in every panel.

Figure 25 – PGS Distribution Plot Figure 25. Score distribution plot (histogram + density overlay) for a binary outcome. The two groups are shown in different colours, and the Mann–Whitney p-value is displayed above the plot.

Percentile forest plot

Enable Show percentile forest plot to display ORs (binary response) or β coefficients (continuous response) per percentile category as a forest plot, using the same model and reference category as the percentile category analysis (Figure 26). The reference category is drawn as a filled diamond fixed at OR = 1 (or β = 0); the remaining categories show the point estimate with 95% CI whiskers. When covariates are assigned the model is adjusted for them, and a subtitle lists them. Requires a response variable to be assigned.

Figure 26 – PGS Percentile Forest Plot Figure 26. Percentile forest plot for a binary outcome, adjusted for covariates. Each row is a score percentile band; the reference band (◆) is fixed at OR = 1.

ROC curve

Enable Show ROC curve to display a Receiver Operating Characteristic curve with the area under the curve (AUC), computed via the trapezoidal rule (Figure 27). The PGS-only curve is always drawn; when covariates are assigned, two further curves are overlaid for comparison — PGS + covariates and covariates only — each with its own AUC. Requires a categorical response: for a binary response one curve set is shown; for a polytomous response each non-reference level is drawn against the reference in its own panel.

Figure 27 – PGS ROC Curve Figure 27. ROC curves for a binary outcome comparing the PGS-only, PGS + covariates, and covariates-only models, each annotated with its AUC.

Calibration plot

Enable Show calibration plot to display a calibration plot (Figure 28). Individuals are binned into deciles of predicted probability from the logistic model; the observed event rate per decile is plotted against the mean predicted probability, with the diagonal marking perfect calibration. A Hosmer–Lemeshow chi-square statistic is annotated. Requires a categorical response (binary, or each non-reference level of a polytomous response).

Figure 28 – PGS Calibration Plot Figure 28. Calibration plot for a binary outcome: observed vs predicted event rate by decile of predicted probability, with the Hosmer–Lemeshow statistic annotated.

When both Weighted and Unweighted scoring modes are active, every plot shows each mode in a separate side-by-side panel within the same figure.

References

Sole X, Guino E, Valls J, Iniesta R, Moreno V (2006). SNPStats: a web tool for the analysis of association studies. Bioinformatics 22(15):1928–1929.

Daniel S, Sinnwell J (2026). haplo.stats: Statistical Analysis of Haplotypes with Traits and Covariates when Linkage Phase is Ambiguous. R package version 1.9.8.3, https://CRAN.R-project.org/package=haplo.stats.

Acknowledgments

This tool has been developed with support of the Instituto de Salud Carlos III (ISCIII), “Programa FORTALECE del Ministerio de Ciencia e Innovación”, through the project number FORT23/00032 and the Consortium for Biomedical Research in Epidemiology and Public Health (CIBERESP), action Genrisk.

This tool was migrated from SNPstats.net using Claude AI.