Results

Benchmarking & Validation

TeChBioT conducted structured benchmarking across biological and chemical domains to validate performance, generate training datasets and establish operational robustness under realistic conditions.

Chemical agents

Several real agents and simulants were applied for CWA benchmarking.

Reference measurements were carried out with:

GC/FID
GC/MS

to confirm retention behaviour and generate chemically interpretable chromatographic fingerprints that could be used as training material for data processing and modelling.

In parallel, the consortium:

Built a library for HT-IMS and HT-GC-IMS
Quantified sensitivity, response and selectivity
Compiled retention- and drift-time descriptors for the selected simulants

Selectivity & Stress Testing

Selectivity was stress-tested under:

Varying humidity
Strong gasoline backgrounds

This exposed:

Drift-time shifts
Additional peaks
Sensitivity losses

while keeping identification feasible.

AI Model Enablement

These benchmarking activities directly fed the initial AI models in two ways.

1. Stable reference inputs

The early model work leveraged GC/FID and GC/MS chromatograms as stable, information-rich inputs to develop and validate preprocessing, alignment and classification strategies before transferring and expanding them to GC-IMS/IMS domains.

2. Variable real-world datasets

The progressively more variable datasets — including mixtures, interferents and outdoor measurements — shaped the first operational AI use-case as rapid alarm screening (target vs non-target) and motivated robust normalisation and peak-handling methods.

The outdoor campaign using the DPM simulant on a UGV then provided the first end-to-end demonstration of AI-supported safe/alarm decisions under field conditions.

Biological agents

Biological benchmarking used:

MALDI-TOF MS as a reference approach for microbial identification and dataset generation to support ML/DL development
Py-GC-MS to understand the analytical consequences of pyrolyzing complex biomass

Initial attempts to pyrolyze bacterial biomass yielded complex, unresolved chromatograms.

To reduce complexity and enhance volatilization, an alternative sample preparation protocol was developed.

A significant fraction of bacterial biomass consists of non-volatile lipids, which are valuable for bacterial identification.

Derivatization Requirement

Analyzing these macromolecules requires derivatization to improve their volatility and stability.

Field-Relevant Workflow Demonstration

Field-relevant biological workflows were demonstrated using:

Water pre-concentration
Recovery steps prior to pyrolysis analysis

The reported filtration-to-pellet workflow illustrates a practical route to:

Increase analyte load
Improve detectability in realistic matrices

The document further notes the robustness and usability of the approach in the mobile-lab setting, including identification in a tap-water matrix.