WO2019138226A1

WO2019138226A1 - A method for screening psychoactive substances using nmr spectrometer having a permanent magnet

Info

Publication number: WO2019138226A1
Application number: PCT/GB2019/050057
Authority: WO
Inventors: David Williamson; Oliver Sutcliffe; Ryan MEWS; Kate KEMSLEY
Original assignee: Oxford Instruments Industrial Products Limited
Priority date: 2018-01-10
Filing date: 2019-01-10
Publication date: 2019-07-18
Also published as: CA3087973A1; DE112019000371T5; GB201800377D0; CA3087973C; GB2571817A; GB201900341D0; DE112019000371B4; NZ766434A; GB2571817B

Abstract

A method of screening for psychoactive substances using an NMR spectrometer having a permanent magnet with a magnetic field strength of from 0.5 to 2.5 Tesla is provided. A measured spectrum of a sample is generated using the NMR spectrometer, the sample comprising a test compound. The measured spectrum is then compared to one or more reference spectra, each said reference spectra relating to a psychoactive substance. The test compound is then categorised into a class of psychoactive substances based on said comparison.

Description

A METHOD FOR SCREENING PSYCHOACTIVE SUBSTANCES USING NMR

SPECTROMETER HAVING A PERMANENT MAGNET

FIELD OF THE INVENTION

The invention relates to a method for screening psychoactive substances using a NMR spectrometer having a permanent magnet with a magnetic field strength of from 0.5 to 2.5 Tesla.

BACKGROUND TO THE INVENTION

There is a need for suspected contraband which have been seized by law enforcement to be quickly and accurately screened to identify the presence of any psychoactive substances present therein. One setting where there is particular need is within prisons, where there has been a dramatic rise in the use of “new psychoactive substances” (NPS) over the last few years. Controlled drugs are currently identified within UK prisons either through the technical analysis of a biological specimen (such as urine, hair and blood) or by using ion mobility scanners. For established substances these techniques are generally effective, however these methods have severe limitations in terms of selectivity (a potential for false positives), sensitivity (in particular when low levels of psychoactive substances are present within a mixture), an ability to identify new or previously unreported psychoactive substances and turnaround time between sample acquisition and identification.

Raman spectroscopy has recently been considered for NPS detection and although portable systems are commercially available, they are generally limited by low sensitivity and an inability to determine unknown psychoactive substances within bulk mixtures.

It is desirable that a psychoactive substance screening process is provided which is easy to implement, for example so that a result may be obtained by the security officer who has seized the suspected narcotic, without the need for a skilled chemist or the acquisition of biological specimens. This process should also be implemented with the use of a small, low-cost apparatus so as to promote its uptake at prisons, police stations, airports and the like. The present invention is set in the context of solving these problems.

Nuclear magnetic resonance (NMR) spectroscopy provides a method of probing the electronic environment surrounding NMR active nuclei and inferring structure of the molecules to which these nuclei are bound. NMR spectroscopy involves placing a sample containing NMR active nuclei (such as hydrogen-1 ) within a static external magnetic field B₀ generated by a NMR spectrometer. A radio frequency (RF) electromagnetic pulse is then applied to the sample by the spectrometer at a frequency which is characteristic to the NMR active nuclei of interest. The energy provided by this pulse is“absorbed” by these nuclei and re- emitted as radio waves at the resonance frequency of the active nuclei. This signal is known as the free induction decay signal (FID) and is detected and analysed by the spectrometer to infer the chemical structure of that sample.

High-field NMR spectrometers are routinely used by chemists, for example when synthesising new compounds, so as to verify the structure of the compound produced or to fully elucidate the structure of an unknown compound. These spectrometers typically require large and expensive superconducting magnetic assemblies, as well as specialist knowledge to operate. Furthermore it can often take several days to identify the structure of a compound using such apparatus. This is because of the number of different NMR experiments that need to be performed to generate sufficient data to fully elucidate the structure.

Low-field“benchtop” NMR spectrometers have recently been developed as a cheaper, more compact and portable apparatus for performing certain types of NMR experiments which may otherwise have been performed in high-field NMR spectrometers. Instead of requiring a dedicated infrastructure, trained staff and extensive installations, benchtop NMR spectrometers use permanent magnets that generate a magnetic field between 0.5 and 2.5 Tesla at the sample holder of the spectrometer. These field strengths can be too low to fully elucidate the chemical structure of an unknown compound in some instances.

Benchtop NMR spectrometers are generally self-contained units which may be placed directly on a laboratory bench or surface and moved as necessary. These spectrometers are typically straightforward to operate by a non-specialist user. An example of such a benchtop NMR spectrometer is the Pulsar™ provided by the Oxford Instruments Group. Although these devices have wide potential utility, their usage so far has been primarily restricted to chemists, for example within the food and drinks industry, and for educational purposes.

SUMMARY OF THE INVENTION

A first aspect of the invention provides a method of screening for psychoactive substances using an NMR spectrometer having a permanent magnet with a magnetic field strength of from 0.5 to 2.5 Tesla, the method comprising executing a computer program to perform each of the following steps:

(a) generating a measured spectrum of a sample using the NMR spectrometer, the sample comprising a test compound;

(b) comparing the measured spectrum to one or more reference spectra, each said reference spectra relating to a psychoactive substance and having been acquired at a magnetic field strength of from 0.5 to 2.5 Tesla; and

(c) categorising the test compound into a class of psychoactive substances based on said comparison.

A benefit thereby provided is that a test compound may be screened for the presence of psychoactive substances and any psychoactive substances found may then be categorised in a quick and user-friendly manner. The results may then be displayed to a user, for example via a computer monitor. The comparison performed in step (b) provides a point of distinction over traditional integration-based NMR techniques which are generally reliant on higher resolution data and manual interpretation. The method can subsequently be implemented (and is implemented) using a low-field NMR spectrometer having a permanent magnet. Such apparatus are commonly referred to as“benchtop” NMR scanners and have associated benefits in terms of ease of use, affordability, size and weight. These apparatus are also sufficiently sensitive so as to detect small quantities of a psychoactive substance (e.g. less than 10 mg), or multiple different psychoactive substances within a sample within a short timescale (e.g. less than 10 minutes).

There has been general a bias amongst chemists against using low-field NMR spectrometers for screening compounds. This is due to concerns that the field strength may not be high enough to obtain a conclusive result for the compound’s chemical structure, particularly where the sample contains a mixture of different compounds. For example, it is common for there to be significant spectral overlap in NMR data acquired at low magnetic field strengths which can make it difficult for an operator to identify a compound according to the spectral features. We have found however that a low-field NMR spectrometer can be used to enable the class of a psychoactive substance to be identified, even if it is not always possible to identify its exact chemical structure to a sufficiently high confidence level. Furthermore although the presence of any spectral overlap may hamper manual interpretation of the data, it generally does not impact the ability to categorise or identify the test compound once the measured spectrum is compared to one or more reference spectra. The ability to rapidly screen for the presence of a controlled substance in-situ, as well as its class, may be highly valuable for a security or law enforcement officer. If further investigation is required to identify the chemical composition of the psychoactive substance, the sample may be sent to a laboratory for full analysis.

The application of benchtop NMR spectrometers for screening psychoactive substances (sometimes referred to as “controlled substances”) is particularly desirable within the field of security and law enforcement since no specialist training is required to operate these devices. Furthermore, because the comparison is performed by a computer, rather than requiring a manual assessment to be made, the screening process is made altogether easier for the layman to perform. This is in contrast to traditional screening processes, including those using high-field NMR spectrometers.

In the context of drug screening, NMR technology provides a particular benefit in that the class of psychoactive substance may be identified even if this substance is newly-synthesised and has not previously been identified. The likely effects of the substance, in particular any psychoactive effects, may therefore be predicted according to the class of psychoactive substance to which it belongs. Appropriate action may then be taken if a person is suspected to have ingested this substance.

The measured spectrum is compared to one or more reference spectra having been acquired at a magnetic field strength of from 0.5 to 2.5 Tesla. Preferably the one or more reference spectra have been acquired at the same field strength as the measured spectrum. Furthermore these spectra are preferably acquired using an NMR spectrometer having a permanent magnet. Most typically the one or more reference spectra have been obtained in advance at a remote location. However the spectrometer is preferably of the same model or type as that used in step (a). Reference spectra acquired using the same test compound as analysed in step (a) may therefore exhibit similar a similar spectral response to the measured spectrum, which assists in the comparison process performed. In contrast, if the measured spectrum were compared to reference spectra having been acquired using a high field NMR spectrometer then manual intervention (e.g. of a trained chemist) may be required in order to perform step (c).

In practice the measured spectrum may comprise a processed NMR spectrum or a processed set of NMR spectra. For example, the measured spectrum may comprise a normalised form of an NMR spectrum obtained for the sample. Most typically step (a) comprises obtaining an NMR spectrum of the sample using the spectrometer and generating one or more shifted spectra from the NMR spectrum, each said shifted spectrum corresponding to the NMR spectrum on a shifted frequency scale. The measured spectrum may therefore comprise the one or more shifted spectra. Each of the one or more shifted spectra may hence take the place of the measured spectrum in the comparison of step (b). Changes in the concentration of the test compound can cause a frequency shift of the resulting NMR spectrum. Although the concentration of the substance(s) used to form the one or more reference spectra may be known, the concentration of the test compound is unknown. The formation of a set of “shifted spectra” therefore increases the likelihood of one of the sample spectra aligning with a reference spectrum, which assists with the comparison process in step (b).

Step (b) preferably comprises the following step: (i) performing a similarity match between the measured spectrum and the one or more reference spectra. For example, in the event that the measured spectrum comprises a set of shifted spectra, a similarity match is performed in step (i) between each said shifted spectrum and the one or more reference spectra. Each said reference spectrum in step (i) is preferably a compound reference spectrum corresponding to an NMR spectrum obtained from a particular psychoactive compound. Alternatively however each said reference spectrum in step (i) may relate to a class reference spectrum exhibiting an average spectral response for compounds within a particular class of psychoactive substances.

It is desirable to test for mixtures of psychoactive substances within the screening process. The similarity match in step (i) is therefore preferably performed between the measured spectrum and a plurality of reference spectra, and step (b) preferably further comprises the following steps: (ii) selecting one or more combinations of the reference spectra based on the similarity match performed in step (i); (iii) calculating one or more mixture reference spectra, each said mixture reference spectrum corresponding to an addition of a respective combination of reference spectra selected in step (ii); and (iv) performing a similarity match between the measured spectrum and the one or more mixture reference spectra. The mixture reference spectra are therefore calculated, rather than being empirically measured. This reduces the number of reference spectra that need be stored in memory and the amount of experimentation required to implement the screening process.

In order to limit the processing time required, the combinations selected in step (ii) may correspond to a subset of the reference spectra compared in step (i), such as those found to have the highest similarity (corresponding to a lowest correlation distance) to the measured spectrum. The comparison process may therefore be adaptive, wherein the output of one step informs the actions taken in the next, and so on. Certain combinations of reference spectra may be disallowed in step (ii) in recognition that the combination of the corresponding substances is not encountered in practice (e.g. synthetic cannabinoids with steroids). This further reduces the computational load. In the event that the reference spectra in step (i) are compound reference spectra, the one or more mixture reference spectra will correspond to mixtures of the corresponding compounds. Alternatively, if the reference spectra in step (i) are class reference spectra, the mixture reference spectra in step (ii) will correspond to mixtures of these classes.

It is particularly desirable to consider different combinations and weightings of mixtures as part of the comparison. For example, the method may further comprise iterating steps (ii) to (iv) using different combinations of selected reference spectra to form the mixture reference spectra in each iteration until a matching spectrum is found to a sufficiently high similarity or until a threshold number of iterations is performed. The different combinations may include different weightings of the reference spectra. In the above iterative process, each iteration may be performed without reference to the last. Although this may be quick to implement and perform, more accurate results may be obtained if the process is adaptive, for example so as to converge on a solution. Preferably therefore the similarity match in step (iv) is performed between the measured spectrum and a plurality of said mixture reference spectra, and step (b) further comprising the following steps: (v) selecting one or more of said mixture reference spectra based on the similarity match in step (iv); (vi) adjusting the proportions of the reference spectra forming said selected mixture reference spectra so as to form one or more adjusted mixture reference spectra; and (vii) performing a similarity match between the measured spectrum and the one or more adjusted mixture reference spectra. Once again it is desirable to reduce the processing time by limiting the number of mixture reference spectra selected. For example, only a subset of the mixture reference spectra calculated in step (iii) may be selected in step (v), such as those having the closest similarity to the measured spectrum. Adjusting the proportions in step (vi) then generates yet further reference spectra for comparison against the measured spectrum, which makes it more likely to identify a matching spectrum to a sufficiently high confidence score. Steps (v) to (vii) could also be an iterative process, which is repeated until a matching spectrum is found to a sufficiently high similarity or up until a threshold number of iterations is performed. Alternatively, step (vi) may comprise performing a non-negative least squares spectral fitting of the measured spectrum onto a subset of matched mixture reference spectra obtained from step (iv). This process is typically computationally more demanding (particularly over large data sets) however it has been found to be more accurate in estimating the correct proportions and components of any psychoactive compounds present in the sample.

As more reference spectra are combined or their weightings adjusted there is a higher likelihood of false positives being returned (wherein the comparison process incorrectly identifies the measured spectrum as corresponding to a particular class or compound of psychoactive substance). It is therefore desirable to apply a penalty factor to the matched mixture reference prior to any comparison. Step (iv) may hence further comprise applying a penalty factor to the matched mixture reference spectra. Similarly, step (vii) may further comprise applying a penalty factor to the matched adjusted mixture reference spectra.

It is also desirable that step (b) further comprises a process of generating a list of the psychoactive substances corresponding to the spectra compared with the measured spectrum in each said similarity match, wherein said psychoactive substances are ranked within the list according to the similarity of the corresponding reference spectra with the measured spectrum based on said similarity matches. This process may form step (viii). Step (c) may then comprise categorising the test compound into a class of psychoactive substances based on the psychoactive substance identified in the list as corresponding to the closest matching spectrum. For example, a correlation or confidence score may be attributed to each of the spectra or the corresponding substances compared against the measured spectrum (including, where appropriate, each shifted spectrum). Step (c) may then comprise a process of categorising the test compound into a class of psychoactive substances based on the psychoactive substance identified in step (viii) as corresponding to the closest matching spectrum.

Various different techniques may be used for any the aforementioned similarity matches. For example, a similarity match may be performed using a distance metric, such as cityblock metric. Other potential metrics that may be used include the Euclidean distance, the cosine distance, and the correlation distance. The correlation distance (defined as 1 minus the Pearson’s correlation coefficient between measurement vectors) between the measured spectrum and each of the reference spectra may be taken and a closest (i.e. most similar) match identified according to which has the smallest correlation distance. These distances may be calculated using the whole spectral region or from smaller, selected spectral regions.

The similarity match may be calculated based on a tolerance parameter, preferably with a tolerance value of ± 0.1 ppm. The tolerance parameter defines the range over which the spectra are‘shifted’ and thus provides for an amount of latitude on the precise positions of spectral features. NMR spectra may undergo a frequency shift at all resonances, typically due to sample concentration effects. The x-axis of an NMR spectrum is given in parts per million (ppm), which is calculated as the frequency of a resonance divided by the base resonance frequency of the magnetic field. The frequency shift is almost negligible for high- field spectrometers where the base resonance frequency is high (e.g. in excess of 400 MHz) however this shift becomes more observable for NMR spectra generated by low-field spectrometers. This frequency shift is therefore preferably corrected for automatically using the tolerance parameter during data analysis.

The comparison in step (b) preferably involves a comparison of the measured spectrum exhibited as magnitude data and the one or more reference spectra. Optionally, the one or more reference spectra are also exhibited as magnitude data for this comparison. The measured spectrum therefore does not need to be phase corrected. Phasing typically requires manual intervention and so the use of data that has not been phase corrected means that steps (a)-(c) may be more readily automated. The screening process may therefore be completed in less time than if the comparison were performed based on phase-corrected absorptive spectra. Magnitude spectral data is often avoided by chemists due to its tendency to exacerbate the overlap between spectral peaks and increase the bases of peaks. Furthermore, spectral peaks cannot be integrated using magnitude spectral data: thereby making a traditional spectroscopic analysis difficult. Step (b) replaces the need for such analysis techniques however such that these effects no longer impact the ability to identify or categorise the test compound.

The categorisation is performed by comparison to one or more reference spectra. The one or more reference spectra may comprise one or more class reference spectra, wherein each said class reference spectrum is an average of a plurality of NMR spectra each relating to a different compound of psychoactive substance within the same class of psychoactive substances. A closest match may therefore be found between the measured spectrum and the class reference spectrum from each class of psychoactive substances, for example. Alternatively or additionally, the one or more reference spectra may comprise at least one compound reference spectrum corresponding to an NMR spectrum obtained from a psychoactive substance of a particular compound. Compound reference spectra may be obtained for the compounds of psychoactive substances most likely to be present in the sample. Reference spectra may also be obtained from non-psychoactive substances (typically those commonly combined with psychoactive substances) for comparison against the measured spectrum.

The sample may contain a plurality of different substances, each of which being represented by the measured spectrum. The signal resulting from any psychoactive substances present within the sample however is referred to herein as the“psychoactive substance fingerprint”. Step (b) typically further comprises identifying a psychoactive substance fingerprint in the measured spectrum, said psychoactive substance fingerprint corresponding to a said reference spectrum. This may be the case, for example, where a “match” is found between a component of the measured spectrum and a reference spectrum.

A particular benefit is provided wherein step (c) further comprises estimating the effects of the psychoactive substance when administered by a user. This process may be carried out automatically once the test compound has been classified by reference to an electronic library or database. This information may provide the user with useful information to enable him or her to act accordingly if a person is suspected to have ingested the substance.

Step (c) may further comprise identifying the test compound based on said comparison. In particular, the name of the test compound may be identified. This may be possible, for example, when the one or more reference spectra comprise named compound reference spectra and a match has been found between one of these spectra and a portion of the measured spectrum. The test compound may then be identified as comprising the matching named compound.

The sample may comprise a plurality of psychoactive substances. In this case step (c) preferably comprises categorising each said psychoactive substance into a class of psychoactive substances. Step (c) may further comprise a process of identifying the relative quantities of the different psychoactive substances present in the sample.

Unlike some existing drug screening techniques, the sample may be obtained without the need for a biological specimen. The suspected narcotic itself (i.e. the test compound) may instead be dissolved in a solvent, such as deuterated dimethyl sulfoxide. This process is often practically easier to manage. The measured spectrum may then be aligned by reference to a characteristic peak occurring at a known frequency, the characteristic peak resulting from the presence of the solvent in the sample. Other solvents may also be used, such as: acetone-d6; chloroform-d; water-d2; N, N-Dimethylformamide-d7; dimethyl sulfoxide-d6; methanol-d4; and trifluoroacetic acid-d. Typically the same solvent is used to form the measured spectrum and the one or more reference spectra. A particular benefit provided by this technique is that previously unidentified psychoactive substances may be detected and classified. For example, the test compound may be a different compound from any of the psychoactive substances used to obtain the one or more reference spectra.

The NMR spectrometer is typically operable from 270 to 300 Kelvin. This temperature in particular relates to the temperature of the magnet contained therein and is generally not possible for superconducting magnets. The need for large and expensive cooling assemblies may therefore be avoided. Preferably an NMR spectrometer having a permanent magnet with a magnetic field strength of from 1.0 to 2.5 Tesla will be used.

Steps (a) to (c) may be carried out by a computer that forms part of the NMR spectrometer or instead is simply in communication with it. These steps may also be implemented automatically. The class of psychoactive substances to which the test compound relates may hence be communicated to the user in a user-friendly manner, without the need for any specialist knowledge, for example relating to chemometrics or NMR.

A particular benefit provided by this method is that the process may be carried out in significantly less time than existing screening techniques. For example, steps (a) to (c) may be performed in less than one hour, preferably less than 20 minutes, more preferably less than 5 minutes.

The length of time required to complete steps (a) to (c) may vary according to the number of acquisition scans performed. A suitable number of acquisition scans may hence be selected balancing the need to provide rapid analysis against the need to obtain a measured spectrum having a suitably high signal to noise ratio. Step (a) preferably therefore comprises performing a number of NMR scans on the sample, wherein the number of NMR scans is calculated according to the weight and/or type of material used to form the sample. This calculation is typically performed automatically based on information provided by the user. A range of different NMR active nuclei may be used. For example, measured spectrum may be a hydrogen-1 , carbon-13, nitrogen-15, fluorine-19 or phosphorus-31 NMR spectrum. Typically the measured spectrum may be a one- dimensional spectrum.

A second aspect of the invention provides a computer program product configured to carry out the method of the first aspect. The second aspect shares similar advantages as discussed in connection with the first aspect.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the invention will now be described with reference to the accompanying drawings, in which:

Figure 1 is a perspective of a benchtop NMR spectrometer connected to a computer for use in accordance with an embodiment;

Figure 2 is a flow diagram illustrating a method of screening a psychoactive substance in accordance with an embodiment;

Figure 3 is a measured spectrum compared with a class reference spectrum according to an embodiment;

Figure 4 is a measured spectrum compared with a compound reference spectrum according to an embodiment; and

Figure 5 is a flow diagram illustrating steps of a screening process in accordance with an embodiment.

DETAILED DESCRIPTION

Figure 1 is an illustration of a system 1 comprising a “benchtop” NMR spectrometer 2 connected to a computer 10 for use in accordance with an embodiment of the invention. The spectrometer 2 comprises a sample holder 5, shown as a cylindrical cavity in the spectrometer 2, and two NdFeB permanent magnets 7a and 7b operable at room temperature. The magnets 7a and 7b have opposing north and south poles such that a static, primary magnetic field B₀ of from 0.5 to 2.5 Tesla is provided at the sample holder 5. In use, a sample (not shown) is placed in a sample holder 5 within the primary magnetic field B₀. Two magnets are not necessarily required and other magnet geometries are envisaged, such as a cylindrical magnet.

The NMR spectrometer also comprises a set of radio frequency (RF) coils 4 operable between 40 to 110 MHz to provide RF magnetic fields perpendicular to the direction of the B₀ field, and a set of B₀ shim coils (not shown) which are used to compensate for inhomogeneities within the B₀ field. The RF coils 4 are operable to detect RF FID signals emanating from a sample within the sample holder 5, as well as to apply RF fields to the sample. Additional features such as heaters, a vacuum chamber and vibration dampeners that are typically found in benchtop NMR spectrometers may be also provided, however these are not illustrated for clarity purposes.

The benchtop spectrometer 2 is a self-contained unit which may be placed directly on a laboratory bench or surface and moved as necessary (unlike superconducting NMR spectrometers). The spectrometer 2 is connected to a computer 10 which is operable to analyse the results obtained by the spectrometer 2. Alternatively, this analysis may be performed remotely by a server (not shown) via an Internet connection. The computer 10 comprises a display for the user to view the results of the analysis.

A method for screening a psychoactive substance using the system 1 of Figure 1 will now be discussed with reference to the accompanying flow diagram in Figure 2 and the spectra of Figures 3 and 4.

A sample is obtained at step 101 by placing 3 grams of a material containing a test compound, suspected to be a psychoactive substance, into a vial containing 1 ml_ of dimethyl sulfoxide (DMSO). The vial is agitated and left to stand for five minutes so as to dissolve substance in the solvent. If present, any remaining solids are then removed from the vial or the liquid solvent is transferred to another container, such as a 5 mm NMR sample tube, using a pipette so as to form the sample. At step 102 the sample is placed in a sample holder 5 of the benchtop NMR spectrometer 2. The magnets 7a, 7b generate a magnetic field B₀ which is highly homogeneous, such that each part of any sample within the sample holder 5 experiences a substantially identical magnetic field strength with a common direction. Once the sample has been stabilised within the NMR spectrometer 2, for example by stabilising the temperature and applying any “shimming” of the magnetic field, an NMR signal is obtained from the calibration sample.

The remaining steps of the method are performed by the computer 10. Each successive step may be executed automatically so that the test compound may be classified in a user-friendly manner. A computer program product, for example a data storage device, is provided comprising code configured to cause the computer 10 and the spectrometer 2 to carry out these steps.

A“90 degree” RF pulse is applied to the sample by the RF coils 4 of the spectrometer 2. This is achieved using a radio frequency signal of 50 MHz which is pre-calibrated with respect to the known magnetic field strength experienced by the sample so as to cause the hydrogen nuclei in the sample to deflect out of alignment with the applied magnetic field and precess about the main field direction. This precession decays and the resulting free induction decay (FID) back into alignment causes a release of energy which is detected as by the spectrometer RF coils 4 as an FID signal. The FID signal is digitised by the spectrometer 2 and communicated to the computer 10 for analysis.

The signal to noise ratio has been found to increase with the square root of the number of times the acquisition is repeated using the applied RF signal (known as“scans”), at least for under 100 scans. At step 103, a suitable number of scans is calculated by the computer 10 based on information provided by the user concerning the weight and/or type of material used to form the sample. More scans will typically be required when only a small quantity of the test compound is present in the sample. In the present embodiment three scenarios are considered. In a first scenario, if the material used to form the sample is a powder or liquid weighing more than 10 mg, 16 scans are required (taking in total approximately 3.5 minutes to perform). In a second scenario, if the material is a tobacco-like smoking material suspected to have a controlled substance sprayed onto it, 64 repeat scans are required (taking in total approximately 14 minutes to perform). In a third scenario, if the material is a powder or liquid weighing less than 10 mg, 128 scans are required (taking approximately 28 minutes). This may be the case for example where only trace amounts of the substance are found on a seized item, such as a container. It should be understood that the analysis is not limited to a specific number of scans and that these examples are purely indicative only. At step 104 the selected number of scans is performed, the resulting FID signals are detected, processed by software on the computer 10 and then combined together into an aggregate FID signal.

At step 105 the aggregate FID signal is converted by suitable frequency analysis software into frequency space so as to obtain a frequency spectrum exhibited as magnitude data. Suitable Fast Fourier Transform software may be used to achieve this in order to reduce the time spent performing this conversion into frequency space.

Traditionally, when the time domain FID signal is Fourier transformed, it produces “real” and “imaginary” components. Ideally to enable simple identification of compounds in the sample, the real components should contain all of the information concerning the absorption spectrum. However, in actual NMR acquisition (where the precession of the hydrogen nuclei does not start in phase with the experimental frame of reference) a phase correction must be applied in order to obtain a pure absorption spectrum as the real output of the Fourier transform. This phase correction is performed by software which adjusts the real and imaginary components of the Fourier transform by different amounts for each sample. However it is notoriously difficult to obtain consistent phase corrections across large data sets, which impacts on the accuracy of the statistical analysis. Manual intervention is therefore often required. In embodiments of the invention however it has been found that the presence of a psychoactive substance in a sample can be identified to a sufficiently high confidence level without the need for phase-corrected NMR data. As a result, it is desirable that no such phasing is performed following the Fourier transform. The measured spectrum exhibited as magnitude data may instead be used for the analysis. The need for manual intervention and time-consuming processing steps is therefore avoided.

The frequency spectrum then undergoes additional processing such that the chemical shift for the spectrum is calculated based on tetramethylsilane as a reference point (as is routine in NMR). A“measured spectrum” is thereby obtained at step 106 wherein the chemical shift data forms the abscissa of the spectrum and the intensity forms the ordinate. The chemical shift data provides information on the particular chemical environments within which the hydrogen nuclei in the sample are located, such that there is a relationship between a particular chemical shift and a particular part of a molecule bearing hydrogen atoms. In the event that the test compound comprises a psychoactive substance, a psychoactive substance fingerprint emanating from this substance will form part of the measured spectrum. The method then proceeds to step 107a and/or step 107b.

At step 107a a region of interest is selected within the measured spectrum for further analysis. In this case the region of interest is a relatively narrow range in which the psychoactive fingerprint is expected to be visible. This region is referred to as the“class-characteristic region”. A class-characteristic region of 0.5-2 ppm is suitable for identifying compounds in the following classes: ephenidines, diphenidines and cathinones. More than one class-characteristic region may be selected in order to identify the presence of a compound from other classes. For example a region of 7-8 ppm may be selected for identifying the presence of any methylenedioxycathinones. Suitable class-characteristic regions can be readily identified for the detection of particular classes of psychoactive substances based upon the known common chemical structure for substances in that class.

Any peaks that do not relate to the test compound, such as those relating to the solvent DMSO, are excluded from further consideration by the selection in step 107a. Then, at step 108a, the class-characteristic region is normalised to correct for variable concentration of test compounds.

The method will then proceed from step 108a to 109a wherein the class of psychoactive substances to which the test compound relates is estimated. This is performed by comparing the normalised class-characteristic region to a plurality of previously obtained class reference spectra stored in memory. Each said class reference spectrum is a mean profile within the class-characteristic region obtained from a plurality of NMR spectra each corresponding to different compounds of psychoactive substances within the same chemical class. For example, a class reference spectrum may be obtained for each of the following psychoactive substance classes: diphenidines, cathinones, ephenidenes, amphetamines and methylenedioxycathinones. These reference spectra are typically obtained at an equal magnetic field B₀ to the measured spectrum (for example using the same spectrometer 1 ), using the same solvent and may be exhibited as magnitude or phase-corrected absorptive data.

Figure 3 provides an example of a normalised, class-characteristic region from a measured spectrum (labelled“Unknown Test Sample”) shown alongside a class reference spectrum (labelled“Average of Ephenidines in Reference Database”). The normalised region of the measured spectrum may be compared to each of the class reference spectra stored in memory and the closest match found. In this embodiment the shaded area represents the cityblock distance between the normalised, class-characteristic region of the measured spectrum and the average of all the known ephenidines in the reference database. The class reference spectra for ephenidines is identified as being the closet matching and so the test compound is classified as an ephenidine.

Steps 107b-109b may also be performed, potentially simultaneously with steps 107a-109a. At step 107b another region of interest is selected from the measured spectrum. In this case the region of interest is a relatively broad range in which the psychoactive fingerprint is expected to be visible, for example 0.5-12 ppm. This region is referred to as the referred to as the“full-fingerprint region”. Any peaks that do not relate to the test compound, such as those relating to the solvent DMSO, are excluded from further consideration by this selection. Then, at step 108b, the full-fingerprint region is normalised to correct for variable concentration of test compounds.

The method will then proceed from step 108b to 109b wherein the specific compounds for any psychoactive substances present within the sample are identified. This is performed using a similarity match wherein the normalised full- fingerprint region is compared to a plurality of previously obtained compound reference spectra stored in memory. Each said compound reference spectra relates to a previously measured NMR spectrum from a sample containing a psychoactive substance of a known compound. These reference spectra are acquired at the same base magnetic field strength and using the same solvent as the measured spectrum. These reference spectra may be exhibited as magnitude data or phase-corrected absorptive data.

The nearest neighbour between the normalised full-fingerprint region of the measured spectrum and each of the compound reference spectra is found using a distance metric. In the present embodiment the correlation distance is calculated. In particular, the cross-correlation is found, restricted to within a certain tolerance range, for the normalised full-fingerprint region of the measured spectrum with each of the compound reference spectra. Correlation distances are thereby obtained for the test compound from each compound reference spectrum. The test compound is then identified as the psychoactive substance relating to the compound reference spectrum with the smallest correlation distance from the normalised full-fingerprint region of the measured spectrum (thus corresponding to the most correlated reference spectrum). Optionally, a threshold distance can be implemented, above which test compounds are deemed to be too dissimilar from the reference compounds. Such test compounds can be reported as“unknown” or“previously unseen”.

Figure 4 provides an example of a normalised, full-fingerprint region from a measured spectrum (labelled “Test compound”) shown alongside the highest correlated compound reference spectrum (labelled“Reference spectrum of 4- fluoroamphetamine”). These two spectra do not exactly align along the x-axis due to concentration effects however this can be corrected for using a tolerance parameter, for example ± 0.1 ppm. In this example the sample was prepared using a test compound relating to a different psychoactive substance than that of Figure 3. In particular, the test compound is identified as 4-fluoroamphetamine. The class of psychoactive substance to which the test compound relates may then be identified or inferred from this information. This method has also been found to be particularly effective for identifying steroids, synthetic cannabinoids and opioids such as fentanyls and heroin.

In the event that the test compound relates to a new, previously unreported psychoactive substance, it may not be possible to identify the compound to which the psychoactive substance relates according to the steps of 107b-109b. In this case it should still be possible to identify a class of psychoactive substance present in the sample according to steps 107a-109a.

The results of steps 109a and 109b are then displayed to a user at step 110, for example via a computer monitor. A confidence score is typically provided alongside each result, in particular if the name of the test compound is identified. The confidence score may be a correlation distance, in which case a lower number indicates a higher confidence score. Optionally, a plurality of likely compounds and/or classes may be identified and ranked according to their confidence score. Furthermore, the effects of the test compound when ingested by a person may be output so as to enable the user, who may be a security or law enforcement officer, or a medical practitioner, to take appropriate action. If the presence of multiple different psychoactive substances is detected within the sample then the respective classes and compound names (if available), may be displayed for each said substance. Optionally, the relative quantities of these substances may also be indicated.

The compound reference spectra and the class reference spectra may each be stored locally within the computer 10 or the spectrometer 2. Preferably however they are stored within a remote database accessed by the computer 10 via the Internet. This enables the database to be continually updated with more reference spectra from a remote location. A further embodiment of the invention will now be described with reference to the flow diagram of Figure 5. This embodiment is directed to a process of identifying the exact compound of a psychoactive substance present in the sample (equivalent to steps 107b-109b from Figure 2) however a similar process may be performed for simply identifying the class of a psychoactive substance within a sample, as will later be described. The following method uses a minimum distance metric to match a measured proton magnitude spectrum of a sample to a database of stored reference spectra. Although this method relies on calculating the correlation between the spectra, other distance metrics could be used.

The method proceeds as set out in steps 101 to 105 of Figure 2. The frequency spectrum then undergoes additional processing such that the chemical shift for the spectrum is calculated based on tetramethylsilane as a reference point, and the full-fingerprint region of the NMR spectrum is selected and then normalised (as before). At this point additional processing is then applied to the frequency spectrum. In particular, at step 201 the spectrum is converted into a“shifted” matrix by circularly shifting the spectrum along the chemical shift axis (i.e. the x- axis in Figure 4) to simulate the frequency-shifting effect from a range of sample concentrations. In general, the concentration of a test compound will not match the concentration of the compounds used to produce the reference spectra. The process of generating a shifted matrix produces a set of different spectra having different axial offsets (corresponding to different potential concentrations) which can then be compared against the reference spectra. This reduces the likelihood of mismatching occurring along the chemical shift (ppm) scale which could otherwise compromise the identification process. These spectra are referred to herein as the“shifted spectra”.

The method then proceeds to step 202, at which point the correlations are calculated between each of the shifted spectra and each of a plurality of compound reference spectra. This process is performed using a collection of direct matrix operations. For each of the compound reference spectra, only the highest correlation value for the set of shifted spectra is retained. Then, at step 203, a single-compound match list is returned, ranking the compound reference spectra (and/or their corresponding compounds) based on descending order of correlation.

The method then proceeds from step 203 to step 204, whereby the ranking order returned from step 203 is used to generate“mixture reference spectra” formed from pairs of compound reference spectra. Notably, the mixture reference spectra are calculated, rather than being experimentally measured, and correspond to possible mixtures of the reference compounds. In particular, the spectral components of different compound reference spectra are added, in this case in proportions of 1 :1 (alternative weightings will later be considered). The depth of search is set by selecting an appropriate number of potential compound combinations, taking into account the available computing power. This number may be varied in accordance with the correlation values output from step 203. For example, if step 203 indicated a high correlation for certain test compounds, only these test compounds may be considered as forming part of the mixture. If step 203 indicated only a low correlation for the different test compounds considered, a higher number of possible mixture combinations may be considered. In practice this could be implemented by considering only those compounds for which the correlation distance was below a threshold value. Alternatively, only a fixed number subset of compounds will be selected in step 204 for forming mixture reference spectra, for example corresponding to the ten lowest correlation distances calculated in step 203. Optionally certain combinations of compounds will not be combined to form a mixture reference spectrum (e.g. synthetic cannabinoids with steroids) in recognition that these combinations are never encountered in practice. This improves the accuracy of the screening process and reduces the processing time.

Next, at step 205, the mixture reference spectra will be normalised. For example, an area-normalisation process may be followed in which these spectra are normalised so that their integrated intensity equals one. Alternatively a “proton-count normalization” process could be followed (where possible) in which the integrated area of each spectrum is set to the number of protons known to contribute to the measured signal. This aids in the identification of mixtures of compounds with very different numbers of protons (e.g. paracetamol has 9 protons and heroin has 23 protons). Alternative potential normalizations could take into account other factors that might affect overall spectral magnitude, such as different compound solubilities.

Correlations of each of the shifted spectra to each of the mixture reference spectra will then be calculated at step 206. Fast matrix handling functions are used to reduce the processing time. Correlations across mixtures (including the many“incorrect” matches) are generally somewhat higher than from the single compounds at step 203. In other words, step 206 is more likely to produce false positives than step 202. An“over-fitting penalty” is therefore applied at step 207 to the correlations from step 206 to compensate for this. Establishing an effective penalty value may be selected based on the properties of the database at hand and, if necessary, empirically adjusted.

The proportions of the compound reference spectra forming the mixture reference spectrum are then adjusted in step 208. In practice this may be achieved by calculating new mixture reference spectra (referred to herein as “adjusted mixture reference spectra”) using different weightings of matching compound reference spectra returned from step 203. Steps 204 to 207 correspond to a first pass through identifying any mixtures in the sample. Their performance reduces the overall processing power and time required to estimate the test compound(s) present in the sample. In particular, the output from step 206 may be used to inform which reference spectra to consider in step 208. For example, adjusted mixture reference spectra will only be formed in step 208 from combinations of compound reference spectra which, when combined in step 204, produced mixture reference spectra that had a correlation distance to one of the adjusted spectra that was below a threshold number. In principle, it would be possible to move step 203 to step 208 but at the cost of a longer analysis time, as step 208 would then be computationally more intensive.

The proportions of the compound reference spectra forming the mixture reference spectra may be adjusted by performing a constrained (non-negative) least squares spectral fitting of the shifted spectra onto a subset of pairs of most correlated compound reference spectra identified in step 203. A fast algorithm is used for this fitting (e.g. as described by‘Fast algorithm for the solution of large- scale non-negativity-constrained least squares problems’, Benthem & Keenan, J. Chemometrics 2005). This method returns regression coefficients that can only take positive values and which broadly indicate the proportions of the different reference compounds present. The correlation of each shifted spectrum with each of the adjusted mixture spectra is calculated. A mixture match list is then returned at step 209, indicating, for each of the adjusted mixture reference spectra generated, the weightings of the psychoactive compounds used to form the adjusted mixture reference spectra, and its correlation to at least one of the shifted sample spectra.

The method then proceeds to step 210, at which point the overall ranked identification list is generated. Here the correlation lists from the single compound and mixture matches output at steps 203 and 209 are concatenated and sorted to generate the overall ranked identification list. The top match (having the lowest correlation distance, as weighted to include any penalty factors applied) is then returned to the user as the identified test compound(s). In the event that this match is a mixture of psychoactive compounds, the relative quantities of these compounds may be estimated based on the weighting of the compound reference spectra used to form the corresponding mixture reference spectrum or adjusted mixture reference spectrum. The class of the psychoactive substance in the sample can then be identified as corresponding to the class of the top matching psychoactive substance.

Steps 201 to 209 can be generalised into three main stages. Stage one (steps 201-203) relates to correlating the measured spectrum with compound reference spectra, stage two (steps 204-207) relates to correlating the measured spectrum to mixture reference spectra, and stage three (steps 208 and 209) relates to optimising the proportions in the mixture reference spectra. The combination of these stages has been found to be particularly effective in terms of accurately identifying any psychoactive compounds present in a sample. In particular, it has been found that this process can identify the compound of any psychoactive substances present in a sample to with a success rate in excess of 90% in less than 10 minutes.

In an alternative embodiment one or more of the steps described above with reference to Figure 5 may be completed in an iterative manner. For example, steps 204 to 207 may be repeated for a pre-set number of iterations or until a mixture reference spectrum is generated for which the correlation distance is below a threshold number (indicating a sufficiently high level of similarity between the mixture reference spectrum and the measured spectrum). Similarly the weightings of the components used to form the mixture reference spectra may be adjusted in an iterative manner: either for a threshold number of iterations or until a reference spectrum is produced for which the correlation distance is below a threshold number.

A similar process to that described above with reference to Figure 5 may be performed to simply identify the class of a psychoactive substance present in a sample. This process could be followed instead of the abovementioned process or in addition to it (for example where a psychoactive compound was not identified in step 210 to a sufficiently high confidence score). In this case, following step 105, instead of selecting the full-fingerprint region of the NMR spectrum, the class-characteristic region is selected. This region is then normalised and a shifted matrix formed (as before). The method will then proceed as described with reference to Figure 5 however class reference spectra will be used instead of compound reference spectrum. Any mixture reference spectra calculated will also correspond to mixtures of classes, rather than compound mixtures. Such a process may be computationally quicker to perform, however its results provides less specificity. Furthermore the process is not appropriate for all classes, as some do not exhibit a characteristic signal. Compounds within these classes are better identified and classified using the earlier described method, which focuses on identification of the compound itself.

In a yet a further embodiment the method may additionally compare the shifted spectra to one or more reference spectra corresponding to non-psychoactive substances. For example, the process of Figure 5 may be extended to consider three-component mixtures, the third component being a non-psychoactive substance, such as a background residual solvent signal. The isolation and, optionally subtraction, of any signals resulting from non-psychoactive substances may assist with correct identification of any psychoactive substance in the sample. However, further over-fitting penalties may need to be applied to compensate for the higher likelihood of false positives.

As will be appreciated an improved method is provided for screening for psychoactive substances. Advantageously, the class of a psychoactive substance present in a sample may be identified in less time than some existing techniques and typically in less than one hour. Furthermore the method is straightforward to operate and sensitive to low levels of psychoactive substances present within a mixture. Furthermore, the method is suitable for identifying the presence of new or previously unreported psychoactive substances. NPS as well as more conventional psychoactive substances, such as heroin, cocaine and steroids, may therefore be detected.

Claims

1. A method of screening for psychoactive substances using an NMR spectrometer having a permanent magnet with a magnetic field strength of from 0.5 to 2.5 Tesla, the method comprising executing a computer program to perform each of the following steps:

(b) comparing the measured spectrum to one or more reference spectra, each said reference spectra relating to a psychoactive substance and having been acquired at magnetic field strength of from 0.5 to 2.5 Tesla; and

2. A method according to claim 1 , wherein step (a) comprises obtaining an NMR spectrum of the sample using the spectrometer, and generating one or more shifted spectra using the NMR spectrum, each said shifted spectrum corresponding to the NMR spectrum at a different frequency shift, wherein the measured spectrum comprises each said shifted spectrum.

3. A method according to claims 1 or 2, wherein step (b) further comprises the following step:

(i) performing a similarity match between the measured spectrum and the one or more reference spectra.

4. A method according to claim 3, wherein each said reference spectrum in step (i) is a compound reference spectrum corresponding to an NMR spectrum obtained from a particular compound.

5. A method according to claims 3 or 4, wherein the similarity match in step

(i) is performed between the measured spectrum and a plurality of reference spectra, and wherein step (b) further comprises the following steps:

(ii) selecting one or more combinations of the reference spectra based on the similarity match performed in step (i); (iii) calculating one or more mixture reference spectra, each said mixture reference spectrum corresponding to an addition of a respective combination of reference spectra selected in step (ii); and

(iv) performing a similarity match between the measured spectrum and the one or more mixture reference spectra.

6. A method according to claim 5, further comprising iterating steps (ii) to (iv) using different combinations of selected reference spectra to form the mixture reference spectra in each iteration until a matching spectrum is found to a sufficiently high similarity or until a threshold number of iterations is performed.

7. A method according to claim 5, wherein the similarity match in step (iv) is performed between the measured spectrum and a plurality of said mixture reference spectra, and wherein step (b) further comprising the following steps:

(v) selecting one or more of said mixture reference spectra based on the similarity match in step (iv);

(vi) adjusting the proportions of the reference spectra forming the selected mixture reference spectra so as to form one or more adjusted mixture reference spectra; and

(vii) performing a similarity match between the measured spectrum and the one or more adjusted mixture reference spectra.

8. A method according to claim 7, wherein step (vi) comprises performing a non-negative least squares spectral fitting of the measured spectrum onto a subset of matched mixture reference spectra obtained from step (iv).

9. A method according to any of claims 4 to 8, wherein step (iv) further comprises applying a penalty factor to the matched mixture reference spectra.

10. A method according to any of claims 7 to 9, wherein step (vii) further comprising applying a penalty factor to the matched adjusted mixture reference spectra.

11. A method according to any of claims 5 to 10, wherein step (b) further comprises generating a list of the psychoactive substances corresponding to the spectra compared with the measured spectrum in each said similarity match, wherein said psychoactive substances are ranked within the list according to the similarity of the corresponding reference spectra with the measured spectrum based on said similarity matches;

and wherein step (c) comprises categorising the test compound into a class of psychoactive substances based on the psychoactive substance identified in the list as corresponding to the closest matching spectrum.

12. A method according to any of claims 3 to 11 , wherein each said similarity match is performed using a distance metric, such as cityblock metric.

13. A method according to any of claims 3 to 12, wherein each said similarity match is calculated based on a tolerance parameter, preferably with a tolerance value of 0.1 ppm.

14. A method according to any of the preceding claims, wherein the comparison in step (b) comprises a comparison of the measured spectrum exhibited as magnitude data with the one or more reference spectra.

15. A method according to any of claims 1 to 3 and 5 to 14, wherein at least one of the one or more reference spectra comprises a class reference spectrum corresponding to an average of a plurality of NMR spectra each relating to a different compound of psychoactive substance within the same class of psychoactive substance.

16. A method according to any of the preceding claims, wherein at least one of the one or more reference spectra comprises a compound reference spectrum corresponding to an NMR spectrum obtained from a psychoactive substance of a particular compound.

17. A method according to any of the preceding claims, wherein step (b) further comprises identifying a psychoactive substance fingerprint in the measured spectrum, said psychoactive substance fingerprint corresponding to a said reference spectrum.

18. A method according to any of the preceding claims, wherein step (c) further comprises estimating the effects of the psychoactive substance when administered by a user.

19. A method according to any of the preceding claims, wherein step (c) further comprises identifying the test compound based on said comparison.

20. A method according to any of the preceding claims, wherein the sample comprises a plurality of psychoactive substances and wherein step (c) comprises categorising each said psychoactive substance into a class of psychoactive substances.

21. A method according to claim 20, wherein step (c) further comprises identifying the relative quantities of the different psychoactive substances present in the sample.

22. A method according to any of the preceding claims, wherein the test compound is dissolved in solvent within the sample.

23. A method according to any of the preceding claims, wherein the test compound is a different compound from any of the psychoactive substances used to obtain the one or more reference spectra.

24. A method according to any of the preceding claims, wherein the NMR spectrometer is operable from 270 to 300 Kelvin.

25. A method according to any of the preceding claims, wherein steps (a) to (c) are performed automatically.

26. A method according to any of the preceding claims, wherein steps (a) to (c) are performed in less than one hour.

27. A method according to any of the preceding claims, wherein the measured spectrum is a hydrogen-1 , carbon-13, nitrogen-15, fluorine-19 or phosphorus-31 NMR spectrum.

28. A method according to any of the preceding claims, wherein the measured spectrum is a one-dimensional spectrum.

29. A method according to any of the preceding claims, wherein step (a) comprises performing a number of NMR scans on the sample, wherein the number of NMR scans is calculated according to the weight and/or type of material used to form the sample.

30. A computer program product adapted to carry out the method of any of the preceding claims.