You have an X value and one or more Y values — e.g. time vs concentration, dose vs response, age vs weight.
Use when: plotting a curve, a regression, a dose-response, or any scatter plot.
Graphs: scatter, line, area, dose-response curves · Analyses: correlation, linear regression, nonlinear curve fitting, Chou-Talalay synergy

Each column is a separate group or condition. Rows are replicates (individual measurements).
Use when: comparing several groups, e.g. Control vs Drug A vs Drug B, each measured independently.
Graphs: bar, box-whisker, dot plot, violin · Analyses: t-test, ANOVA, Mann-Whitney, Kruskal-Wallis

Like Column, but with a second grouping factor. E.g. Treatment (rows) × Time point (columns).
Use when: you have a two-way design — two independent variables, each with multiple levels.
Graphs: grouped bar, stacked bar, column bar · Analyses: two-way ANOVA, Friedman

Tracks when events (death, relapse, failure) occur. Each row is a subject; columns are time and event status (1=event, 0=censored).
Use when: analysing time-to-event data with some subjects still alive/event-free at the end.
Graphs: Kaplan-Meier curve · Analyses: log-rank test

A frequency table — rows and columns are categories, cells are counts (not continuous measurements).
Use when: asking whether two categorical variables are associated — e.g. treatment group vs outcome (responder/non-responder).
Graphs: pie, bar · Analyses: chi-square, Fisher's exact test

A dose-response grid: Drug A concentrations across columns, Drug B concentrations down rows. Each cell is % inhibition.
Use when: testing whether two drugs interact synergistically or antagonistically across a concentration range.
Graphs: dose-response heatmap, 3D Bliss surface · Analyses: Bliss independence model

Compares the means of two independent groups.
Why use it: The workhorse of biology. Use it when your two groups are separate populations (e.g. control mice vs treated mice) and the data is approximately normal.
Assumes: Normal distribution, roughly equal variance (or use Welch's correction, which is the default here).
Example: Is cell viability different between untreated and drug-treated wells?

Like the unpaired t-test, but each value in group 1 is linked to a specific value in group 2 (same subject, before/after).
Why use it: Pairing removes between-subject variability, giving you more statistical power. Use it when measurements are matched — the same animal measured pre- and post-treatment, or the same experimenter running both conditions.
Example: Blood pressure in the same patients before and after a drug.

The non-parametric alternative to the unpaired t-test. Compares medians instead of means — doesn't assume normality.
Why use it: When your data is skewed, contains outliers, or your sample size is too small to assess normality reliably (n < 8–10 per group).
Example: Pain scores (1–10 scale), tumour sizes with extreme outliers, or any ordinal data.

Non-parametric equivalent of the paired t-test. Tests whether the median difference between paired observations is zero.
Why use it: Paired data that's not normally distributed, or when you have ordinal paired measurements.
Example: Ranked preference scores from the same participants under two conditions.

Tests whether the means of three or more independent groups differ. Always follow up with a post-hoc test (Tukey HSD is recommended) to find which groups differ.
Why use it: Running multiple t-tests inflates your false-positive rate. ANOVA controls for this by testing all groups simultaneously.
Assumes: Normality, equal variance (homoscedasticity), independence.
Example: Comparing growth rates across 4 different media conditions.

Tests the effect of two independent factors and their interaction. E.g. does treatment effect depend on time point?
The interaction term is often the most interesting result — it tells you whether the two factors amplify or cancel each other.
Example: Testing three drugs at four time points — does efficacy change over time, and does that change differ between drugs?

Non-parametric one-way ANOVA. Ranks all values and compares groups on their rank distributions.
Why use it: When your data violates normality assumptions and you have ≥3 independent groups. Dunn's post-hoc test is used for pairwise comparisons.
Example: Severity scores (mild/moderate/severe coded 1-3) across multiple treatment groups.

Non-parametric equivalent of repeated-measures ANOVA. Use when the same subjects are measured under ≥3 conditions and data isn't normal.
Example: 10 patients rated on 4 different drugs — ordinal preference scores.

Measures the strength and direction of a linear relationship between two continuous variables. r = +1 is perfect positive, r = −1 is perfect negative, r = 0 is no linear relationship.
Assumes: Both variables are roughly normally distributed and the relationship is linear.
Example: Does protein concentration correlate with absorbance?

Measures monotonic (consistently increasing or decreasing) relationships. Ranks the values first, then correlates. Works on ordinal data and non-normal distributions.
Example: Does a clinical score (1-10) correlate with biomarker level, even if neither is normally distributed?

Fits a straight line (Y = mX + c) to your data and gives you the slope, intercept, R², and confidence intervals. Unlike correlation, regression implies a directional relationship (X predicts Y).
Example: Standard curve for ELISA — use absorbance to predict concentration.

Fits models like the 4-parameter logistic (4PL) dose-response curve, IC50/EC50 curves, or exponential decay. The 4PL model is the standard for drug dose-response: it gives you IC50, Hill slope, top, and bottom plateaus.
4PL formula: Y = Bottom + (Top−Bottom) / (1 + (IC50/X)^HillSlope)
Example: Fitting a drug inhibition curve to find the half-maximal effective concentration.

Tests whether there is an association between two categorical variables in a contingency table. Compares observed counts to what you'd expect if the variables were independent.
Requires: Expected counts ≥5 in most cells. For small samples, use Fisher's exact test instead.
Example: Is treatment group (drug vs placebo) associated with response category (responder/non-responder)?

Like chi-square but exact — no approximation assumptions. Required when any expected cell count is <5. Standard for 2×2 tables in biology.
Example: 8 animals — 3/4 treated survived, 1/4 control survived. Is this significant?

Kaplan-Meier estimates the probability of surviving past each time point, accounting for subjects who were lost to follow-up (censored). The log-rank test compares survival curves between groups.
Censoring: A censored observation means the event hadn't occurred by the end of follow-up — the subject left the study or the study ended. This information is still valuable and KM handles it correctly.
Example: Tumour-bearing mice — does treatment extend survival?

Uses the median-effect equation to calculate a Combination Index (CI) at each dose. CI < 1 = synergy, CI = 1 = additive, CI > 1 = antagonism. Requires XY data for Drug A alone, Drug B alone, and the combination.
Example: Two cancer drugs tested individually and together across a dose range — are they synergistic?

For a full dose-response matrix (Drug A × Drug B). Calculates the expected inhibition if the drugs act independently: E_expected = E_A + E_B − E_A×E_B. The Bliss score = observed − expected. Positive = synergy, negative = antagonism.
Visualised as: a colour-coded heatmap and a 3D surface plot — green peaks are synergy, red valleys are antagonism.
Example: 6×6 concentration matrix of two drugs, each from 0–100 nM.

Shapiro-Wilk (small samples, n<50) and D'Agostino-Pearson (larger samples) test whether your data is consistent with a normal distribution.
Caution: With very small n (<8), these tests have low power — they often miss non-normality. With very large n, they detect trivial deviations. Use them as guidance, not gospel.
Rule of thumb: If p > 0.05, proceed with parametric tests. If p < 0.05, consider a non-parametric alternative.

Grubbs' test identifies a single outlier in a dataset — the value furthest from the mean — and tests whether it's significantly more extreme than expected by chance.
Important: Only run this test once per dataset. It is not designed to be run iteratively. If a data point is flagged, investigate the cause before removing it — an outlier might be a pipetting error, or it might be the most biologically interesting result.
Alpha (α): The significance level. Default 0.05 means a 5% chance of falsely flagging a non-outlier.

Mean, median, SD, SEM, CV%, min, max, and quartiles for each column. Always look at these first — before running any test — to understand what your data looks like.
SD vs SEM: SD describes the spread of individual values. SEM describes the precision of the mean estimate. Bars in graphs are usually SD (variability) or SEM (precision). Never use SEM to imply the spread of your data is small.

Apply log (log₁₀, ln), square root, reciprocal, or Z-score transforms to selected columns. Useful for:
• Log transform: making skewed data more normal, or converting concentrations for dose-response analysis
• Z-score: normalising columns to mean=0, SD=1 for comparison or PCA
• Square root: stabilising variance in count data
Access via the right panel or the Change menu.

Reduces many variables to a smaller set of uncorrelated components that capture the most variance. PC1 explains the most variance, PC2 the next most, and so on.
Why use it: To visualise high-dimensional data (e.g. proteomics, gene expression), spot clusters, or identify which variables drive the most variation.
Tip: Z-score your columns before PCA if they're on different scales (e.g. mixing mg/mL with absorbance units).
Example: 20 cytokine measurements across 50 patients — do treated and untreated patients cluster separately?

P-values tell you whether an effect is real — effect sizes tell you how big it is. Cohen's d (for t-tests) and η² (for ANOVA) are standard.
Cohen's d: 0.2 = small, 0.5 = medium, 0.8 = large.
Why it matters: A p < 0.001 with d = 0.1 means a tiny but statistically detectable effect — probably not biologically meaningful. A p = 0.06 with d = 0.9 means a large effect with insufficient power to reach significance.

Use the Format axes button (graph toolbar) to set axis titles, min/max ranges, and tick intervals. Click the style sidebar (right side of graph view) to change colours, marker size, line thickness, and error bar style.

Significance brackets: Run an analysis first, then look for the bracket tool in the graph toolbar to add significance annotations directly to the figure.

Error bars: SD shows the spread of your data. SEM shows the precision of the mean. 95% CI is the most interpretable for most audiences. Choose based on what you want to communicate.

PNG — raster image, best for presentations and web. Export at high resolution for publications.
SVG — vector format, scales to any size without pixelation. Best for figures that will be resized, or edited in Illustrator/Inkscape after export.

For multi-panel figures, use ⊞ Layout to arrange graphs on a canvas, then export the whole layout as one image.

If you run 20 t-tests, you expect 1 false positive by chance even if nothing is real. Use ANOVA + post-hoc tests instead of running many t-tests. Post-hoc tests (Tukey, Dunn's) correct for multiple comparisons automatically.

SEM is smaller than SD and makes graphs look "tighter." But it describes the precision of the mean estimate, not the actual spread of your data. In most biological contexts, SD is more honest. Always label your error bars.

An outlier test tells you a value is statistically unusual — not that it's wrong. Before removing any data point, check your lab records for a documented error (wrong dilution, pipetting mistake, equipment fault). If there's no documented cause, report the analysis both with and without it.

p = 0.049 and p = 0.051 are not meaningfully different. Report exact p-values. Also report effect sizes — a statistically significant result with a tiny effect size may be biologically irrelevant. A non-significant result with a large effect size may be limited by sample size.

A p-value is not the probability that your hypothesis is true. It is the probability of seeing a result at least as extreme as yours, assuming the null hypothesis is true.

p < 0.05 does not mean "the result is real." It means "if there were no effect, we'd see something this extreme less than 5% of the time." That's meaningful — but it's not certainty. Replication, effect size, and biological plausibility all matter equally.