From instrument export to results.

The problem

In most labs I walk into, analysing a study and writing it up takes days. Someone opens each plate in Excel, copies the numbers next to the layout, normalises to the vehicle wells, drags a curve fit across the doses and pastes the answer into a report. Every analyst does it slightly differently, transcription errors creep in, and the controls are checked by eye, if at all.

The fix is to describe the assay once, in a layout file and a settings file, and let one pipeline run the same steps every time.

What you get

• An automated analysis pipeline. Raw plate exports in, results out. Each plate is normalised and judged against your own settings: signal floor, replicate spread, edge effect, Z′. A four-parameter curve is fitted to every compound and cell line and the IC50 reported, with every number traced back to the file it came from.
• Control charts built in. Each run's controls land on a chart with limits calculated from your last 30 good runs. A failed run is flagged on the day and drops out of future limits, so one bad day never moves the baseline. Absolute limits and a minimum Z′ are yours to set.
• Reports that write themselves. A Word report populates from the run, in your template, with the standard sections filled and the plates, curves and QC verdicts included.
• The first piece of a lab brain. The pipeline writes to one on-site database that the control charts, study log and reports read from. Inventory, equipment and time tracking plug into the same store later.

The demo on this page is a prototype on synthetic data. The version I build in your lab is fitted to your assay, your instruments and your report format.

How it works

1. Drop in the exports and the layout. In the demo, that is sixteen plate exports and one file saying which well holds what: compound, dose, replicate, control.
2. Press Run. Each plate is read, the layout applied and the values normalised to the vehicle wells. Every plate gets a verdict against your settings: signal floor, replicate spread, edge effect, Z′. A four-parameter curve is fitted to each compound and cell line, and the IC50 is reported.
3. Check the controls. Each run's controls are judged against limits from the lab's own previous runs, using two simple rules: one point beyond three standard deviations is a fail, two in a row beyond two is a warning.
4. Report and trace. A Word report populates itself, and the run log shows which input file every number came from.

When the next study changes, you change the layout file, not the code. The film shows a compound dropped, a third replicate added and a lysis control introduced, all by editing one file.

What a build involves

1. Describe the assay. Half a day with whoever runs it today: plate layout, controls, normalisation, acceptance criteria, and what the report needs to say.
2. Build and compare. I set up the pipeline on a machine in your lab and run it side by side with your current analysis on past studies until the numbers agree.
3. Hand over. Your team runs it; I document the settings, train whoever owns it, and stay on call for the first few studies.