NIRLab

A Streamlit app for exploring near-infrared (NIR) spectra exported as multi-block JCAMP-DX (.jdx) files. It takes a raw export through a standard chemometrics workflow, from raw spectra to PLS regression, without writing code.

The analysis logic lives in nir_core.py, independent of the interface, so it can be reused from scripts or another front end. The interface text is in Italian.

Workflow

The app has one tab per step:

1. Raw spectra — load .jdx files; each block becomes one sample, exported as a wavelength × sample matrix.
2. L, a, b — CIE Lab colour values stored with each block (##CONCENTRATIONS=), if present.
3. Outliers — distance of each SNV spectrum from the mean, to spot suspicious samples.
4. Duplicates — reproducibility of repeated measurements (an original and a bis replicate).
5. SNV — Standard Normal Variate correction on the final dataset, after excluding samples and averaging duplicates.
6. PCA — scores and loadings of the SNV spectra, with points optionally coloured by L, a or b.
7. Clustering — Ward hierarchical clustering with dendrogram and silhouette score.
8. PLS — regression of the SNV spectra on a chosen target variable, with leave-one-out cross-validation to pick the number of components.

Tables can be downloaded as CSV from each tab.

Run

pip install -r requirements.txt
streamlit run app_streamlit.py

On Windows: py -m streamlit run app_streamlit.py.

Data organisation

Code and data are kept apart. Each analysis is a project under data/, created from the sidebar:

data/<project>/
├── raw/            source .jdx files
├── <group>/        CSVs and figures for each sample group
├── LAB_values.csv  files shared across groups
└── reports/        write-ups

Samples are grouped by the alphabetic prefix of their name (Sal12 → sal), and a trailing bis marks a duplicate of the sample with the same number (Sal7bis).

data/ is excluded from version control: spectra and results stay on your machine.

Input format

The parser expects multi-block JCAMP-DX files where each block has a ##TITLE=, an ##XYPOINTS=(XY..XY) section and, optionally, ##CONCENTRATIONS= with L, a, b values. It was developed on exports from a BUCHI NIRWise instrument (diffuse reflectance, 400–1700 nm). Files from other instruments may need small changes to the parser.

Requirements

Python 3.10+ and the packages in requirements.txt (streamlit, pandas, numpy, scipy, scikit-learn, matplotlib, plotly).

License

MIT