Dimensionality reduction & clustering / Trend (expression pattern) analysis
Your data stays in your browser: files are read and processed on your device, never uploaded to any server, and cleared when you close the page. Free, no sign-in.
Each gene is first z-scored across conditions (mean 0, SD 1), keeping only the shape of its profile, and k-means then groups genes with similar shapes into trend clusters.
The first chart shows each cluster’s mean profile; the second has one block per cluster, with the cluster’s genes as grey lines (evenly thinned when there are many) and the cluster mean in colour.
With k = 0 the tool picks the k between 2 and the upper limit with the highest mean silhouette width; this is a guide only, and you can set k yourself.
Column order defines condition order. With replicates, upload a sample–condition table: replicates are averaged per condition and conditions follow their first appearance in that table.
Trend clustering is not a statistical test; it is advisable to select changing genes first (differential analysis or variance filtering).
Hartigan & Wong (1979) Appl Stat 28:100–108 (k-means); Rousseeuw (1987) J Comput Appl Math 20:53–65 (silhouette). z-scores use the n − 1 standard deviation; reported centres, within-cluster sum of squares and silhouette widths can be recomputed independently from the partition (they match scikit-learn). With many genes and gradual patterns there are many near-equivalent local optima: on the example data the within-cluster sum of squares is within 0.1% of scikit-learn KMeans (n_init = 100) and the adjusted Rand index of the partitions exceeds 0.95; more restarts lower the sum of squares further.
Up to 30,000 genes (above 3,000 genes the silhouette is computed on a seeded sample of 3,000).
Send us your data and research question and you will receive a written plan within 1 working day: analysis steps, parameter rationale, deliverables and timeline. Quoted per project.