Start something new
A workflow walks you through a job step by step and is saved in AWS as you go. Leave it at any step, start another, and come back to it here, from any computer. Everyone at Analysis Zero can see and open saved workflows.
Saved workflows
Latest runs
Ladder protocols
A protocol says why a ladder is run, what is measured and how. Plan it here, approve it, and give its code to the uploader on the instrument laptop. The first scan measured under a version freezes it; a change after that is saved as the next version, so every scan keeps the plan it was measured under.
Find a run
Every run in AWS, newest first. Click one to open it. To find a single file by name anywhere in AWS, use "Bring in files" on the Add spectra tab.
Ladder protocol
No protocol attached to this batch. Attach one to see which samples have been measured.
Results by sample
Average of each sample's scans, leaving out any scan that is flagged. A sample is the start of the file name, before the scan number.
Files, newest first
Click a file to see its spectrum and details.
Spectrum
Pick a file on the left.
One point per scan, drawn as soon as it is predicted: about 10 to 30 seconds after the scan ends on an instrument laptop sending to AWS, a few seconds in the desktop app. A scan gives one result when it ends; nothing is predicted while it is still being measured. Blue points are readings that passed, and the line joins them. A hollow orange point is a reading its own model flagged, or a scan whose spectrum check found a problem: don't report it. Another model's flag on the same scan does not change the colour. Drag across a chart to zoom in on a stretch of time; double-click a chart to see the whole run again. This tab checks every 3 seconds while it is open.
Air checks
A scan whose name starts with "Air" is taken as a scan of the empty flow cell and is checked as it arrives: is the cell clean and dry? A clean cell shows only the probe window's line near 417 cm⁻¹. Water, ethanol or lipid left behind show on top of it. Figures are a percent of the window line. The limits come from this week's air scans on one instrument, so a flag means "look at this scan", not proof.
Replay earlier scans as if live
Plays earlier scans into a new run, one at a time, in the order they were measured. Each is predicted on arrival and drawn above, so you can rehearse or demo a run. Replay runs start REPLAY- and are never copied to SharePoint or sent to a customer.
Try a model on files
Run up to five models on files you pick, from AWS or from this computer, and see the predictions. This is a test: no run's recorded results change, and the model in use stays the same. A model from a different matrix than the files is allowed and marked, which is how you check whether it carries over.
Models (up to 5)
Look at spectra
Every scan at full resolution (1 cm⁻¹ steps, 301 to 2000 cm⁻¹), as measured or processed. Pick a batch, tick the scans to show, and choose a view.
Spectra
Hover to read values. Drag across the chart to zoom in; double-click it to zoom out. Click a scan in the table below to pick it out.
Scans shown
Check a set of spectra
Run this on a ladder before building a model, or on new spectra to see how they compare. It checks each spectrum for saturation, cosmic-ray spikes and weak signal, says which processing steps are needed, runs a PCA to find scans that do not match their repeats, and can look for peaks that the reference spectra do not have.
Batches
Levels for coloring and for the analyte checks come from the batch's ladder protocol.
Spectra
Pick one batch above to see its spectra. Darker lines are higher levels when the batch has a protocol.
Drag across the chart to zoom in; double-click it to zoom out.
PCA scores
Each dot is one scan. Repeat scans should sit together. A red ring marks a scan to check.
What each component is made of
Model library
Every model, grouped by the matrix it was trained in. A Raman model holds only for its own matrix, so a scan is predicted only with the model in use for its batch's matrix and analyte. Switching models does not change results already recorded. Use "Run again with current models" on the Results tab for that.
Peak-ratio readings
For a substance that is a large part of the sample, such as ethanol in a process buffer. The reading is the substance's own peak divided by a signal from the rest of the sample in the same scan (for ethanol: its line at 879 cm⁻¹ against the rise of water's band above 3100 cm⁻¹), turned into a level by a curve fitted to samples of known level. A cloudy sample or a change in laser power lowers both signals together, so the ratio holds. It is not a model: it has no check that a scan looks like the training scans, only a flag when the ratio is outside the calibrated range. One in use reads every new scan of its matrix, and its results show on the Live and Results tabs beside the models'.
Matrices
The fixed list of sample matrices: the process step, buffer or formulation the analyte is measured in. Batches and models pick from this list, so one matrix never gets two spellings. Only an admin adds to it.
Hossein's model library
The 12 CQA models Hossein built: the lipids in the lipid stock, ethanol and sucrose in drug product, mRNA and the NTPs in IVT, and mRNA in the DS pool. Each is rebuilt here from his own data file and settings and checked against the numbers his notebook printed; a model file is never loaded. They appear in the library above under their unit operation as the matrix, not switched on. They were trained on another program's data, so they stay in AWS and are never copied to a customer's folder. Only Analysis Zero and Orange Antennae sign-ins see them.
Build a new model from a ladder
Pick the batch (or batches) holding the ladder spectra. If the batch has a ladder protocol with the measured reference values filled in, no sheet is needed. Otherwise upload a sample sheet with a Sample_name column that matches the start of the file names, a Sample_Number column, and one column per analyte.
The model is for one matrix: the one its batches are set to, or the one you pick here. It is used only for scans from that matrix.
Batches with the ladder spectra
Add new samples to a model
The new samples must be from the model's matrix. The new version is kept only if it does better on the new samples and no more than 10% worse on the old ones. Otherwise it is saved as a candidate and the current version stays.
Batches with the new spectra
Recent model jobs
Is the model right? Ask the advisor
Pick a ladder whose protocol has the prepared levels and the measured reference values (for example NMR). The advisor checks the spectra, tries each existing model for the analyte in the ladder's matrix as it is, then tries adding these samples to the best one, then a new model, and says which works, why, and what to do next. Its decision still needs a chemometrician's sign-off before results go to a customer.
Ladder batch
Add spectra by hand
Use this if the uploader on the instrument laptop is not running. Each file is fingerprinted in your browser before it is sent, and files already in a batch are never overwritten. Pick the matrix the samples are in, so they are predicted with that matrix's models.
Automatic uploads from a Raman instrument laptop
Set this up once on each computer that runs a Raman. After that nothing is typed: the uploader starts with Windows, watches the Spexware folder, and every scan is sent to AWS on its own, checked, and shown on the Results and Live tabs, usually 10 to 30 seconds after the scan ends. A copy also stays on the computer, and nothing is lost if the network drops.
Once per computer
- Copy the program. Unzip the Raman computer zip straight into C:\, so you get C:\ramanmodel. Do not use a OneDrive folder. Nothing is installed and no administrator rights are needed.
- Double-click Set up Raman computer. An Analysis Zero admin types the upload key from AWS Secrets Manager (UploaderKeySecret) when it asks; it is checked at once. The key never goes in email or chat. It finds C:\MM Data (it asks only if Spexware saves somewhere else), makes the uploader start with Windows, and starts it in a minimised window.
- Test it. In Spexware, save one scan into a new folder inside C:\MM Data. It should appear on the Results tab within a minute.
Every run
- In Spexware, make a new folder inside C:\MM Data for the run and save its scans there. Each folder is one run. A folder called EtOH ladder becomes the run LIT-R1-EtOH-ladder-20261113-01. Scans saved loose in C:\MM Data become one run per day (LIT-R1-scans-<date>-<NN>).
- Pick the run's matrix next to the run on the Results tab. Scans are predicted only with that matrix's models.
Scans saved after the uploader first started on a computer are sent; to send earlier ones too, double-click Send older scans on that computer and give a date. A scan AWS already has is never sent again. A folder inside a run's folder is its own run. Run names matter: a run starting exactly LLY-PoC- is sent on to the Lilly tenant in the Analysis Zero platform. A folder whose name starts with Lilly or LLY is sent as a Lilly run by itself. For other folders at Lilly, double-click Use Lilly run names once; back at Littleton, Use Littleton run names. A folder used before the switch starts a new run, so earlier scans never move into a Lilly run.
The same thing typed by hand, for troubleshooting
NMR files
The NMR computer has its own uploader. Unzip the NMR computer zip straight into C:\ and double-click Set up NMR computer: it takes the same upload key, asks once for the folder the NMR software saves into, makes the uploader start with Windows, and starts it. After that nothing is typed: each folder directly inside the NMR data folder is one run (LIT-NMR-<folder>-<date>-<NN>). Every file of every experiment is copied, fingerprinted, and sent to AWS under s3://<data bucket>/nmr, folders kept.