Review and Correct Results by Hand

Build an app where you check automatic results and correct them by clicking, with dose-response curves and outlier wells as the example.

Automatic analysis gets most things right, but a scientist still has to look: a bubble in a well, a blurry image, a cluster with the wrong label. An app lets you see each result, fix it with a click, and save the corrected version, with a record of what you changed.

Why you'd use it

Catch what the algorithm can't. Edge effects, bubbles and debris are obvious to the eye and easy to miss in a table.
Fix it in seconds. Click a well, an image or a cell to exclude or relabel it; the result updates at once.
Keep an honest record. Save what you excluded with the results, so a reviewer can see how each number was made.

Example: dose-response curves with a bad well

A screening lab runs a cell viability screen (for example CellTiter-Glo) and fits IC50 curves for each compound. One well at a low dose has a bubble and reads far too low, which drags the curve. In the app, the user sees the point sitting off the curve, clicks it to exclude it, and the curve and IC50 refit. The saved table lists the excluded well.

Prompt to build the app

Build a dose-response app. The user uploads or picks a plate-reader export and a plate map with the columns well, compound and concentration in µM. Normalize each plate to the DMSO wells (100%) and the staurosporine wells (0%), fit a four-parameter logistic curve per compound, and show a table with the IC50, its 95% confidence interval and the Hill slope. Highlight wells that look like outliers, and let the user click a well to exclude it and refit. Let them download or save ic50_results.csv, including the excluded wells, and the curve figure as PDF. Add a Load example button.

Review cell type labels:

Build an app where the user picks an annotated .h5ad from their vault, sees the UMAP colored by cell type with the top markers for each cluster, accepts or renames each cluster's label, and saves the reviewed labels as cluster_labels_reviewed.csv with a column for what was changed.

Triage variant calls:

Build an app where the user picks a VCF from their vault or uploads it, sees each variant with its depth, allele fraction and gene, marks each one as keep or reject, and saves the kept variants and a log of the rejected ones.

Steps for anyone using the app

Load your data: Upload from device or Pick from Drylab vault.
Scan the results. Points or images the app flags as likely outliers are highlighted.
Click anything that's wrong to exclude or relabel it. The result updates straight away.
Save the corrected results, with the list of what you changed. See Save App Results Back to Your Vault.

Tips

Name your control wells in the prompt (for example "DMSO in columns 1 and 12"), so normalization is right from the first plate.
Ask the app to flag likely problems (outlier wells, low-quality images), so reviewers know where to look first.
Always save the exclusions. A results table without the list of excluded points can't be reproduced.