Skip to main content
03.07.2024 | כז סיון התשפד

Beating Drugs at Hide and Seek

Machine Learning-Based Forensic Platform to Identify Drugs by Their Physical Properties

Beating Drugs at Hide and Seek

Law enforcement agencies already face many psychoactive substances, but new drugs are constantly being developed and quickly distributed worldwide. These new substances are often hard to identify and may not yet be included in various countries' illegal drug lists. The increasing variety of dangerous drugs challenges monitoring and alert systems.

Professor Hanoch Senderowitz from Bar-Ilan University's Chemistry Department is part of a consortium developing a forensic platform to efficiently identify substances during investigations by characterizing their physical properties. This proposal won an EU grant for fighting crime and terrorism.

The model is based on the understanding that the current challenge isn't just finding dangerous substances during an investigation, but identifying a given substance as illegal, especially when it's a new type of drug.

Existing analytical tools have updated "libraries" for quick identification of dangerous drugs, but this approach has several main problems. Substance libraries aren't regularly updated. Devices struggle with the "matrix effect" - non-target substances mixed with the active ingredient confuse the system. New drugs are distributed and sold in small doses, making them hard to detect and identify. Standard forensic analysis methods are no longer effective in identifying these dangerous drugs or their metabolites due to lack of information on the drug's chemical structure and absence of known reference materials.

The consortium's platform offers a solution to these problems. Samples can be examined using various analytical methods to obtain their spectrum. Each sample has a unique spectrum, acting as its "fingerprint." For isolated and known substances, there are libraries of spectra. The challenge arises with samples containing mixtures of known and new substances, and mixtures of substances with body fluids.

The consortium's idea is to use machine learning models. Researchers will present the computer with a series of spectra of known substances in various forms: isolated, in solutions, and mixed with different body fluids. They will train it to identify psychoactive substances within the disorder.

The new platform, called NARCOSIS (Non-tArgeted foRensic multidisCiplinary platfOrm for inveStigatIon of drug-related fatalitieS), is multidisciplinary, up-to-date, and open to future updates. Its main features include identification tools for use at crime scenes and in labs, a database of substance spectra that can be shared between organizations, AI-based tools for spectrum management, and implementation of machine learning toolkits to help the EU Early Warning System quickly identify new dangerous drugs and respond accordingly.