A method and system for identifying microbes

Abstract

The disclosure relates to a computer-implemented method suitable for identifying or classifying microbes, the method for use with a system comprising a cartridge for receiving a sample, the cartridge

Description

This disclosure relates to a computer-implemented method and system for identifying and / or classifying microbes present in a sample. Background In laboratory settings it is important to be able to identify and classify microbes. As a result, many different techniques have been developed to help with this task. For example, there are a number of phenotypic tests which can be used to observe microbial characteristics for identification and classifications. Such tests include observations of biochemical reactions (for example, reactions with hydrogen peroxide, or substrate utilization tests), Gram staining, and chromatography tests. However, such tests can be rather subjective and are typically time-consuming. To complement or replace established phenotypic methods, genotypic identification may be used. These may be DNA-based techniques for identifying specific microbes in a sample, for example by amplifying and sequencing a DNA extract. However, in addition to the length of time required to obtain satisfactory results, these techniques are typically expensive. The present invention seeks to address these and other disadvantages encountered in the prior art. Summary An invention is defined in the independent claims. Optional features are set out in the dependent claims. Figures Specific embodiments are now described, by way of example only, with reference to the drawings, in which: Figure 1 depicts a system according to the present disclosure; Figure 2 depicts a method according to the present disclosure; Figure 3 is a bar chart depicting a dataset used for training a classification model; Figure 4 is a chart depicting a LIME result of variable weightings; Figure 5 shows an example decision tree that may be used by a classification model; Figure 6 is a confusion matrix showing of the output of a trained classification model; and Figure 7 is a graph depicting receiver operating characteristic curves of a trained classification model. Detailed Description General At its most general, the present invention provides a means for identifying or classifying microbes based on signals received from a plurality of electrochemical transducers arranged to interrogate a sample. In a first example, a computer-implemented method is provided that is suitable for identifying or classifying microbes, the method for use with a system comprising a cartridge for receiving a sample, the cartridge comprising an array of wells, each well comprising an electrochemical transducer. The method may comprise interrogating each of the electrochemical transducers, wherein interrogation of the electrochemical transducers comprises, for each well, generating a plurality of data series by repeatedly measuring a respective plurality of independent variables. The method may further comprise determining, based on the plurality of data series from each well, an identification or classification of a microbe in each well. As used herein, the term "identification" may mean specifically naming a microbe species which is present or dominant in a sample, and "classification" may mean associating a microbe which is present or dominant in a sample with a particular group or classifying it as either belong to a selection of one or more species, or not belonging to the selection of one or more species. Such a method may provide accurate microbial identification or classification, as the electrochemical transducers enable the generation of a wealth of data which can be used for the identification or classification process. This method thereby allows identification or classification to be performed quickly and accurately. In examples, the method may be used to identify a specific microbe in a sample, may be used to distinguish a microbe in a sample between several microbe species, and / or may be used to classify whether a microbe in a sample is of one or more specific species or not. Optionally, the determining may be performed using a trained classification model. The classification model may be a decision tree-based model, for example. For example, the trained classification model may be trained using a gradient boosting algorithm. For example, a classification model may be trained by populating the array of wells with one or more known samples, such that the classification model learns to associate particular dependent variables with the one or more known samples. Optionally, the plurality of dependent variables may be individually weighted and / or scaled for the trained classification model. For example, the individual weightings or scalings may be determined based on training of a classification model. Optionally, interrogation of the electrochemical transducers may comprise, for each well, generating the plurality of data series by repeatedly measuring the respective plurality of dependent variables while controlling at least one dependent variable. For example, the at least one independent variable may be selected from voltage, frequency, and time. The plurality of independent variables may be selected from current, capacitance, impedance, a first derivative of impedance, a second derivative of impedance, phase, and voltage, for example. Optionally, the plurality of dependent variables may further comprise the concentration of microbial matter in the sample, and / or the change (optionally including the rate of change) in the concentration of microbial matter in the sample over time. The concentration of microbial matter may be determined according to a numerical model, for example. For example, the numerical model may be a numerical model as described in UK patent application number 2307287.9, the contents of which are incorporated herein by reference in their entirety. Optionally, determining an identification or classification of a microbe in each well may comprise identifying an organism species. For example, the method may determine a specific organism species that is present in each well. Optionally, determining an identification or classification of a microbe in each well may comprise differentiating between different organism species. For example, the method may determine a specific organism species that is present in each well, selected from a predetermined list of organism species. Optionally, determining an identification or classification of a microbe in each well may comprise determining that an organism species present in a well does not belong to a predetermined list of organism species. In a second example, a system is provided that is suitable for identifying microbial growth. The system may comprise a cartridge for receiving a sample that may comprise a microbe, the cartridge comprising an array of wells, each well comprising an electrochemical transducer. The system may comprise a processor configured to perform a method according to the first example. Optionally, the array of wells may comprise a control subset of wells and a first subset of wells containing a first microbial sample. For example, the control subset of wells may contain no microbial sample. Optionally, the array of wells may comprise a second subset of wells containing a second microbial sample. In this way, the system may be used to determine an identification or classification of one or more microbes present in multiple samples. It will be appreciated that the array of wells may comprise further subsets of wells containing further samples. Optionally, the array of wells may comprise a third subset of wells comprising an antimicrobial agent (which may be referred to herein as an antibiotic agent). For example, susceptibility to the antimicrobial agent may be used by the method (e.g., by a trained classification model) to identify or classify a microbe that is present. It will be appreciated that the array of wells may comprise further subsets of wells containing further antimicrobial agents. In a third example, a computer readable medium is provided, comprising computer-executable instructions which, when executed by a processor, cause the processor to perform a method according to the third example. In a general overview, once a sample has been introduced into each of the wells, the electrochemical transducers in each well are interrogated. A number of scans are performed sequentially in each cell. A first variable is measured over time to generate a first data series. A second data series is generated by additionally measuring either the first, or a second dependent variable over time. Suitable dependent variables to measure during the electrochemical interrogation process include current, capacitance, impedance, a first derivative of the impedance, a second derivative of the impedance, phase, and voltage. Therefore, to generate an example first data series, the frequency of a signal applied to each electrode may be controlled while monitoring changes in impedance. To generate an example second data series, the frequency of the signal applied to each electrode may be controlled while monitoring changes in capacitance. There may be additional data series generated during the interrogation process. This multivariate and multivariable approach generates a wealth of data which can be used (e.g., as the input to a classification model) to determine an identification of a microbe within the sample (e.g., whether a particular microbe species is present or not). The multivariate and multivariable approach which allows the time before a final determination is reached to be 2 hours, or less. Due to this significant reduction in testing time, methods of the present application are extremely valuable as a decision support tool to be utilised by healthcare professionals at the point of care. The methods allow a sample to be interrogated in real time. The presently disclosed method and system therefore enables doctors and other healthcare professionals to access vital diagnostic information. The skilled person will appreciate that the present methods are suitable for any use cases, for example biohazard monitoring for quality control in drink manufacture, water quality monitoring, or identification of a microbe in a biological sample (e.g., a bodily fluid, such as blood or urine). System Figure 1 depicts a system 100 according to the present disclosure. The system 100 comprises a cartridge 110 and a controller, which may be described as a control unit, 120. The cartridge 110 may be referred to as a test cartridge. Depending on the specific implementation, the cartridge 110 may have several purposes, including to securely contain a sample (e.g., a sample from a patient, such as a blood sample or a urine sample, or any suitable bodily fluid); to filter the sample for impurities, for example via alOOpm filter; to contain a growth media and, optionally, a pH buffer and / or antibiotic agent(s); and to facilitate electrochemical interrogation of the sample via the control unit 120. The cartridge 110 comprises a plurality of wells llla-g. The wells llla-g may be described as electrode-sensor wells. Each of the wells llla-g comprises a transducer, in particular an electrochemical transducer. The electrochemical transducers allow electrochemical interrogation of a sample by an array of electrodes. Each electrode-sensor well comprises an independent transducer, which in turn comprises (not depicted in figure 1) a working electrode (WE), a counter electrode (CE), and a reference electrode (RE). Each electrode-sensor well llla-g is physically separated from other cells (i.e. there is no electrode-sensor well 111 situated downstream from another electrode-sensor well 111). While only seven wells llla-g are depicted in the schematic depiction in figure 1, depending on the implementation it should be appreciated that the cartridge 110 may comprise an array of many more of such wells. Each electrode well comprises its own gel layer. The gel composition for a particular well will depend on the usage to which the cartridge is being applied, and which subset or 'configuration'" of wells the particular well falls into. For example, when the cartridge will be used for microbe identification and / or classification, there are three main cell configurations: Blank cell or negative control (no antibiotic agent, but bacteriostatic agent or bacteria filtering is present) Uninhibited growth cell (no antibiotic agent, no bacteriostatic agent or no bacteria filtering) Inhibited growth cell (antibiotic agent present, no bacteriostatic agent or no bacteria filtering). In the simplified implementation depicted in figure 1, for example, it might be that well Illa is a blank cell, in which there is no antibiotic agent present in the gel layer. A bacteriostatic or bactericidal agent is present. Signals and data series generated from this well will be used during pre-processing methods. Wells 111b,c may be uninhibited growth cells, in which the gel layer comprises no antibiotic agent, no bacteriostatic agent and no bacteria filtering means. The remaining wells llld-g are inhibited growth wells. The gel layer of wells Uld,e may comprise a first antibiotic (AB) agent, and the gel layer of wells lllf,g may comprise a second, different AB agent. It will be appreciated that test cartridges according to the present disclosure may be tailored to the particular microbe they are configured to test for (e.g. urinary tract infections, respiratory tract infections, fungal infections etc.). For example, using the example described above, the first AB agent incorporated within wells llld,e may be an agent typically suited for a number of fungal infections, whereas the second AB agent incorporated within the wells lllf,g may be another antibiotic agent particularly suited for a number of other fungal infections. Introducing a patient sample to each well of the cartridge and employing the methods of the present disclosure (described below), it is possible to determine what microbe is present in the sample based on the observed response to the antibiotic agents, in combination with variables observed in the wells. Suitable antibiotics to be featured on the cartridge include: Amoxicillin (AMX) Ciprofloxacin (CIP) Fosfomycin (FOS) Nitrofurantoin (NIT) Trimethoprim / Sulfamethoxazole (TRS). The concentration levels of the antibiotics used on the cartridge may be any of: Breakpoint concentration for the antibiotic in question One dilution down from breakpoint (half of breakpoint concentration) One dilution up from breakpoint (double of breakpoint concentration). Upon introduction of a patient sample (e.g. a blood sample or a urine sample, or any suitable bodily fluid) into the cartridge 110, the sample is distributed through channels into each of the wells containing gel-modified electrode sensors. As soon as the sample reaches the gel present in each well llla-g, bacterial growth may occur, which is monitored through electrochemical changes in parameters such as impedance in a manner which will be described in detail below. At a high-level, the present methodologies may be used to determine an identification or classification of one or more microbes present in the sample based on the sensed parameters. The device 100 comprises electronic communication means 115, which enables communication with the cartridge receiving means 1240. Signals can be passed between the cartridge 110 and the control unit 120 to enable the electrochemical transducers on the cartridge to interrogate the wells 111a-111g, and pass the resulting information back to control unit 120. The system 100 also comprises a controller, in particular a control unit 120, which may take the form of a bench-top instrument. An advantage of the present system is that the control unit 120 is designed to sit on the bench at the point of care, and be positioned as dose to the patients' bedside as possibie. Figure 1 depicts one specific implementation of a control unit 120 according to the present disclosure, but the skilled person will appreciate that a control unit according to the present disclosure may take many forms and that not all of the components shown in figure 1 are essential. Further, while only a single control unit 120 is illustrated, the term "controller" shall also be taken to include any collection of machines (e.g., computers) that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein. Depending on the use case and specific implementation, the purpose of the control unit 120 is to house the cartridge 110, drive the test protocol (sample processing &interrogation), perform measurement(s), and provide an interface with the operator (for example a clinician). The control unit 120 comprises cartridge receiving means 1240, which comprises at least an opening or aperture sized, positioned and otherwise configured for receiving the cartridge 110. The cartridge receiving means 1240 further comprises cartridge reading means configured to interact with the electronic communication means 115 on the cartridge. The control unit 120 comprises one or more processors 1202. The processor(s) 1202 are configured to send control signals in order to interrogate the electrochemical transducers on the cartridge 110, as well as to receive and process signals received from those transducers, in a manner that will be described in detail below. The one or more processors 1202 may comprise both a microcontroller (MC) and a session border controller (SBC). This enables the software residing on the control unit to follow a split architecture of microcontroller (MC) and session border controller (SBC), in order to separate safety critical software components from non-safety critical ones. This structure has the added benefit of the non-safety critical software to be configurable to suit connectivity needs without impacting the development of the safety critical aspects. The micro-controller handles safety critical software. The purpose of the micro-controller is to drive the basic functionality of the control unit 120 (mechanical, thermal and electronics parts). This includes: Self-checks relating to the hardware components; Controlling the Analogue Front End (AFE); Running data processing; Regulating temperature in the wells; and Executing the processing logic (e.g. instructions 1222) for performing the algorithms, methods, operations and steps discussed herein. The control unit 120 further comprises, a main memory and / or a static memory 1204. For example, a main memory may comprise read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc. A static memory may comprise flash memory, static random access memory (SRAM), etc). The control unit 120 may further comprise a secondary memory (e.g., a data storage device 1218). The one or more processors 1202 communicate with each type of memory via a bus 1230. The computing device 1200 may further include a network interface device 1208. The computing device 1200 also may include a video display unit 1210, an input device 1212 (e.g., a keyboard or touchscreen), a cursor control device 1214 (e.g., a mouse or touchscreen), and an audio device 1216 (e.g., a speaker) for providing feedback to an operator of the control unit 120. Components 1210, 1212, and 1214 may be combined to form a graphical user interface (GUI). The data storage device 1218 may include one or more machine-readable storage media (or more specifically one or more transitory or non-transitory computer-readable storage media) 1228 on which is stored one or more sets of instructions 1222 embodying any one or more of the methodologies or functions described herein. The instructions 1222 may also reside, completely or at least partially, within the main memory 1204 and / or within the one or more processors 1202 during execution thereof by the control unit 120, the main memory 1204 and the processor 1202 also constituting computer-readable storage media. Accordingly, disclosed herein is a computer readable medium comprising computer-executable instructions which, when executed by a processor, cause the processor to perform any steps or features of the computer-implemented method disclosed herein. The system 100 further comprises a potentiostat. The potentiostat of the system may be suitable for any of electrochemical impedance spectroscopy (EIS) measurements; Cyclic Voltammetry (CV); linear Sweep Voltammetry (LSV); Staircase Voltammetry (SCV); Normal Pulse Voltammetry (NPV); Differential Pulse Voltammetry (DPV); Square-Wave Voltammetry (SWV); and the like. As such, the potentiostat is capable of analysing, for example, electrical impedance as a function of test frequency, i.e. the potentiostat comprises an impedance analyser. The potentiostat is capable of real-time data capture to generate one or more data series. Fluctuations in impedance indicate variations in microbial growth over time. As the population and metabolic activity of the microbes rise, the resistance of the substance to current flow, which depends on frequency (i.e., impedance), also increases. It is believed that the balance of charged particles within the first substance changes as the microbes proliferate and metabolize the components within it. Therefore, tracking changes in impedance over time provides insight into the rate of microbial growth, which may be used in the present method to identify or classify the microbe which is present or dominant in a sample. Method Figure 2 is a flowchart depicting a method 200 according to the present application. The method is suitable for being performed by the one or more processors on the control unit described above in respect to figure 1. In particular, the method 200 may be performed when a cartridge 110 comprising a sample is inserted in the control unit 120. The sample may comprise a pathogen. Before the method 200 is carried out, a sample is introduced to the cartridge. For example, the sample may be a urine sample from a patient who is suspected to have a UTL In other examples, the sample may be any other suitable bodily fluid, such as a blood sample. The sample is introduced into each of the wells of the array of wells on the cartridge. At the first block 210, the method 200 comprises generating, for each well of the array of wells, a first data series by repeatedly measuring a first dependent variable. At the second block 220, the method 200 comprises generating, for each well of the array of wells, a second data series by repeatedly measuring the first, or a second, dependent variable. Each well of the array of wells on the cartridge comprises an electrochemical transducer, and generating the first and second data series is accomplished via a process of interrogating each of the wells, either sequentially on at the same time. For each data series generated as part of this interrogation process, the dependent or 'measured' variable may be any of current, capacitance, impedance, impedance' (i.e., the first derivative of impedance with respect to time), impedance" (i.e., the second derivative of impedance with respect to time), phase, and voltage. The data series may be generated by repeatedly measuring the relevant dependent variable while controlling an independent variable, such that the first data series is generated at block 210 by repeatedly measuring a first dependent variable while controlling a first independent variable, whereas the second data series may be generated by repeatedly measuring either the first, or the second dependent variable while controlling either the first, or a second, independent variable. The dependent variables may be any of current, capacitance, impedance, a first derivative of the impedance, a second derivative of the impedance, phase, and voltage. The first and second independent variables are selected from voltage, frequency, and time. This is shown in table 1 below: Independent variable Unit Range Dependent variable Unit Voltage V -0.1 to 0.75 Current pA Frequency Hz 0.1 to 100000 Capacitance F Frequency Hz 0.1 to 100000 Current pA Frequency Hz 0.1 to 100000 Impedance O Frequency Hz 0.1 to 100000 Impedance' O Independent variable Unit Range Dependent variable Unit Frequency Hz 0.1 to 100000 Impedance " O Frequency Hz 0.1 to 100000 Phase 0 Time s Oto 5 Voltage V Table 1 The table shows example combinations of independent (controlled) variables and dependent (measured) variables, and the range over which the scanned variables may be scanned. In an example, a single independent variable of frequency and a single dependent variable of impedance may be used to generate both the first and the second data series. Each electrochemical transducer in the array of wells is interrogated, and the impedance is regularly measured with a signal frequency of 0.1Hz to generate the first data series. Similarly, the impedance is regularly measured with a signal frequency of 1Hz to generate the second data series. While the method 200 refers only to a first and a second data series, this is done to aid quick understanding of the invention and to enable brevity of description. The disclosed method may comprise repeatedly scanning, for example, the independent variable through a plurality of values and recording the dependent variable at each value, and in this way generating a plurality of data series. Extending the above example to an implementation which uses five data series xi-x5, an implementation of the present method may repeatedly measure: Xi = voltage X2 = capacitance X3 = impedance' x4 = cell growth in the presence of an antibiotic x5 = impedance" For particular parameters, such as capacitance and impedance and derivatives thereof, multiple readings may be obtained at different frequencies in order to provide more data series for the classification model. Other parameters (such as voltage) may be sampled over time, and a reading at a particular time may be selected as particularly suitable for identifying or classifying a microbe (e.g., it may be found that a voltage reading taken at 4 seconds into a 5 second long measurement period is particularly suited for identifying or classifying a microbe). The cell growth rate in the presence of an antibiotic may be measured according to methods as described in UK patent application number 2307287.9, the contents of which are incorporated herein by reference in their entirety, for example. At a high level, the interrogation of the transducers may involve the sending of control signals, from the control unit 120, to the plurality of electrochemical transducers comprised on the cartridge 110. These control signals enable the control of the independent variable(s). Measurement signals are then returned to the control unit 120 which are indicative of the measured dependent variable(s). The interrogation / measurement proceeds at every electrochemical cell present on the system's cartridge 110 in an independent manner. The measurements may run, for example, sequentially in every independent cell in the same order. At block 230, the method comprises determining, based on the first and second data series from each well, an identification or classification of a microbe in each well. The determination may involve making use of the pro-processing, machine learning / numerical models, models, and expert system techniques for example. In some examples, the determination may involve calculating other factors, such as whether there has been microbial growth in any of the wells of the array of wells, and / or the growth rate of microbes. In addition, in certain examples, determining an identification or classification of a microbe may include determining whether a pathogen present in the sample is susceptible or resistant to a particular antimicrobial agent. The wells of the array of wells may be prepared such that they fall into one of various subsets: a control subset in which the wells contain no microbial sample (e.g., only a buffer solution or growth gel or the like; this may also be referred to as a baseline subset as it provides a baseline reading against which values from other subsets may be compared); a first subset in which the sample will be placed; optionally a second subset in which the sample will be placed and in which a antimicrobial agent will be placed; and optionally a third subset in which a second sample may be placed (e.g., where a second sample is also to be tested for microbe identification and / or classification). The number of wells in each subset may be any number of wells, and there may be a single well in each subset. Additional subsets may comprise further samples and / or further antimicrobial agents. When the cartridge is prepared in this way and a sample potentially containing a pathogen is introduced to the array of wells, then repeated interrogation of the wells as described with respect to blocks 210 and 220 generates a plurality of data series for each well. A subset of those data series are associated with the control subset of wells, a subset of those data series are associated with the first subset of wells, a subset of those data series are associated with the second subset of wells, and so on. By processing these data series in the manner described herein, an identification and / or classification of a microbe present in each subset of wells is obtained. Methods of the present disclosure will now be described under the following sections: 1. Signal processing: to provide clean signal for the models (e.g. smoothing, normalising, etc.) 2. Model(s): predict states based on input data (e.g. numerical fitting or condition checking); Model training and development, and model application; 3. Growth factor evaluation / expert system: interprets outputs from the models (e.g. voting system, rule-based system, etc.) 4. Display of results Signal (pre-)processing There are several pre-processing methods, which provide a framework to standardise experimental output. The objective of the pre-processing step is to provide data in a form which makes comparison between different experiments, instruments, samples, etc. possible, without necessitating a recalibration step to ensure expected outcomes. The pre-processing steps are not essential, but are advantageous, and may include: Normalisation of each trace by division of the entire trace by first measurement value (time to=Os) (see Equation 1) Subtraction of a mean baseline trace (average of all present baseline readings present on the cartridge from all other well types) Each measured variable is considered independently for the purpose of pre-processing. E.g., impedance traces at frequency 100kHz are analysed separately from frequency traces at 10kHz, etc. Equation 1: xt = xt = 0 where t=l is a timepoint at time 7', and t=0 is a first data point acquired within an experiment. Accordingly, each data series associated with a particular well may be pre-processed by dividing the entire data series by the first measurement of that data series as part of a normalisation process. In an implementation in which the baseline subtraction pre-processing step is used, the data series produced by the baseline subset of wells, which may be a single baseline well in which growth gel is present but there is no sample or antimicrobial agent (e.g. no antibiotic agent). This process reduces noise and improves the accuracy of the results. Pre-processing may also include performing data integrity checks to ensure that the data is able to be properly used by a classification model. Data integrity checks may include: Handling missing data (e.g., excluding wells for which data could not be recovered, for example due to equipment malfunction or the like); Categorical variable encoding (e.g., one-hot encoding may be used to prevent the classification model from incorrectly assuming inherent values as ordinal); and / or Numerical variable scaling (e.g., LIME (local interpretable model-agnostic explanations) may be employed for combining and scaling variables, and / or individually weighting the dependent variables for a trained classification model). Model, model training and development An aim of the system is to provide a decision support tool for a healthcare professional (HCP), informing the HCP which microbe is present in a sample (or providing an indication whether one or more of a plurality of predetermined microbes is present or not). The results are generated by an on-board algorithm interrogating the traces derived from electrochemical methods, treated by pre-processing steps described above, and analysed using one or more classification models. Finally, an expert system is tasked to compile the results from the models into a final result shared with the operator (e.g., the HCP). The classification model may be a trained classification model, for example. In a particular example, the classification model may be a decision tree-based model. An example decision tree of such a model is described below with respect to Figure 5. The decision tree-based classification model can be trained using a gradient boosting algorithm. For example, the XGBoost model may be used, which sequentially creates decision trees with each tree correcting errors from its predecessor. Regularization parameters used by XGBoost reduce the influence of individual features (lambda) and encourage sparsity (alpha), ensuring that the model remains efficient. XGBoost optimizes the objective (fitness) function shown in equation 2. Equation 2: Obj(e) = Z”=1L(y;,y;) + where L(yi,yi) is the logistic loss function, which measures the discrepancy between the actual label (e.g., identification or classification) and the predicted probability), guiding the model during training by penalizing errors, and Q(f) is the regularization, controlled by lambda (L2 regularization) and alpha (LI regularization). The regularization restricts the number of leaves and the magnitude of leaf weights in the decision tree to help prevent overfitting, ensuring that the classification model generalizes effectively to new, unseen data. For example, parameters for running XGBoost may be set as identified in Table 2 below. Parameter Name Value Description max_depth 9 Maximum depth for decision tree. Increasing this value will make the model more complex. Zero means no limit. learning_rate 2.64e-l Controls the step size at which the optimizer makes updates to the weights. A smaller eta value results in slower but more accurate updates, while a larger eta value results in faster but less accurate updates. n_estimators 210 Controls the number of trees in the model. Increasing this value may improve model performance, but may also lead to overfitting. min_child_weight 2 The sum of the weights in the child need to be equal to or above the threshold set by this parameter. gamma 3.39e-4 Regularisation parameter. subsample 9.79e-l Controls the fraction of observations used for each tree. A smaller subsample may result in less complex models. coolsample_bytree 9.99e-l Controls the fraction of features used for each tree. A smaller value may result in smaller and less complex models, which may help prevent overfitting. reg_alpha 2.06e-8 The LI regularization term on weights. Larger values means more conservative model, which may help to reduce overfitting by adding a penalty term to the loss function. regjambda 9.16e-6 The L2 regularization term on weights. Larger values means more conservative model, which may help to reduce overfitting by adding a penalty term to the loss function. use_label_encoder false Encoder not used. eval_metric logloss Metric used for monitoring performance during training and for early stopping. LOGLOSS equals negative log-likelihood. Table 2 Bayesian optimization may also be employed for hyperparameter tuning 5 The model training / fitting stage may comprise performing method steps 210 and 220 from figure 2, including any of the steps mentioned above with respect to figure 2. In other words, the training process may comprise inserting a cartridge 110 comprising wells housing one or more samples with a known microbe identifications and classifications into the cartridge-receiving machine 120 (or control unit). The contents of each well can be used for data labelling in training the classification model. Then, for each well of the array of wells, a scan is performed over a first scan variable to generate a first data series, and a second scan is performed over either the same (first), or a second, scan variable to generate a second data series. In this way, training data is collected from samples with a known microbe identification. This training data can be used to train a classification model such that, when presented with a corresponding first and second data series for a sample with an microbe classification, the identification or classification can be estimated (or 'determined') using the trained classification model. For example, Figure 3 shows a dataset comprising 2 clinical studies amounting to around 70 experiments. A mixture of bacteria were contained in the acquired samples, with a single dominant organism acquired for each sample. Figure 3 shows the breakdown of encountered organisms in each study. This dataset was used for training a classification model as described above, with the samples labelled accordingly. Figure 4 is a chart showing the variable scalings obtained after training a classification model using LIME, as described above. Figure 4 shows the following parameters (reading from top to bottom on the left hand axis): a voltage (the open circuit potential, OCP) measured at 4 seconds (OCP_Time_Potential_l_4); a voltage measured at 3 seconds (OCP_Time_Potential_l_3); a voltage measured at 3.5 seconds (OCP_Time_Potential_l_3.5); a capacitance measured at le5 Hz (EIS_Frequency_Capacitance_l_100000); a first derivative of impedance measured at le3 Hz (EIS_Frequency_lmpedance'_6_1000); an impedance measured at 20 Hz (EIS_Frequency_lmpedance_l_20); a capacitance measured at 10 Hz (EIS_Frequency_Capacitance_l_10); cell growth in presence of an antibiotic, amoxicillin (Antibiotic_AMX); a second derivative of impedance measured at le4 Hz (EIS_Frequency_lmpedance"_5_10000); and a second derivative of impedance measured at le3 Hz (EIS_Frequency_lmpedance"_5_1000). Note that EIS stands for 'electrochemical impedance spectroscopy'. That is, after collecting data from a known set of samples (e.g., where the contents of each well are known and labelled, according to the dataset shown in Figure 3), the trained model identifies that the parameters shown in Figure 4 are suitable for identifying or classifying the microbes present. In particular, the LIME results show that a first derivative of impedance measured at le3 Hz (EIS_Frequency_lmpedance'_6_1000); a capacitance measured at 10 Hz (EIS_Frequency_Capacitance_l_10); cell growth in presence of an antibiotic, amoxicillin (Antibiotic_AMX); and a second derivative of impedance measured at le3 Hz (EIS_Frequency_lmpedance"_5_1000) are positively correlated with microbe identification or correlation, and the remaining parameters are inversely correlated with microbe identification or correlation (though are still suitable for making such an identification or correlation). In some examples, the trained model may compared each measured parameter with a threshold value to provide a binary choice of threshold met or not met ('Yes' or 'No'), or the parameter itself may be a binary choice (e.g., cell growth may either be present or not present). For example, the threshold value may be 0. Model application Once the classification model has been trained, the classification model can be used to predict / determine a identification or classification of a microbe for a new sample. A sample is introduced to the cartridge and the electrochemical transducers are interrogated. A plurality of data series is produced for each well (for example, relating to the variables described above in Table 1). This process is repeated continuously. As data is collected for each well over time, the trained classification model can be used in real time to determine a classification or identification of a microbe for each well. During training of a classification model, decision trees (an example decision tree is shown in figure 5 and described below) are constructed sequentially, with each tree learning from the residual errors of the previous one. In examples, each tree may comprise layers of binary splits, where each node represents a decision threshold for a specific feature. The splits may be based on electrochemical data points, such as impedance and capacitance (or any suitable parameter as shown in Table 1), with each branch representing different pathways to classify microbial species. The final nodes may assign class labels, indicating the model's prediction for each sample. This process builds a predictive pathway tailored to microbial differentiation, suited for diagnostic validation. The decision tree may begin with a root node that represents the entire dataset. Here, the model may assess a key feature, such as an electrochemical measurement (e.g., impedance at a specific frequency), and apply a threshold test to determine the first split. For example, "If impedance >X, go left; otherwise, go right." Each split may lead to further nodes, where new feature thresholds may refine the decision. These nodes represent intermediate stages in the tree. For example, in a left branch, the model might assess capacitance at a certain potential, splitting further based on "If capacitance <Y, go left; otherwise, go right." At each node, the model may use thresholds based on specific feature values that best separate the classes, building pathways that move closer to a decision. The process may continue until reaching leaf nodes, which are the endpoints of the decision tree. Each leaf node may represent a final identification or classification outcome, assigning the sample to a specific bacterial class based on the pathway taken. For instance, a path might classify a sample as E. coli based on distinct patterns of impedance and capacitance values. In examples, two approaches were tested: Approach 1: Identify if the organism species is E.coli or other; Approach 2: Identify if the organism species is E.coli, E.faecalis, K.pneumoniae or other. Using the trained classification model as described above, Approach 1 resulted in following correct discrimination: - E.coli 99% - Other 98.5%; and Approach 2 resulted in the following correct discrimination: - E.coli 90.5% - E.faecalis 88.0% - K.pneumoniae 86.3% - Other 95.3%. Figure 5 shows an example of a decision tree 500 which may be used by a trained classification model to provide an identification or classification of a microbe in a sample in a given well. In particular, the decision tree 500 may be used to provide a classification that a microbe is either E.coli or is other (such as in Approach 1). The decision tree 500 is used by the trained classification model to assess each well independently, to provide an indication of the microbe present in the well. The decision tree 500 starts at step S502, which may be during a data collection period or after data has been collected from a plurality of wells as described above. At step S504 the voltage measured at 4s is compared with a threshold value (for example, the model may assess whether the voltage at 4s is less than or equal to 0) and if the value meets the threshold test (e.g., if the voltage measured at 4s is less than 0) then this may be an indication that a classification is not possible or is unknown and the model proceeds to step S524. If the value does not meet the threshold test (e.g., if the voltage measured at 4s is greater than 0) then the model may proceed to a next decision at step S510. Similarly, the voltage measured at 3s is compared with a threshold value at step S506 (for example, the model may assess whether the voltage at 3s is less than or equal to 0), and the voltage measured at 3.5s is compared with a threshold value at step S508 (for example, the model may assess whether the voltage at 3.5s is less than or equal to 0). If the determination at step S506 is that the voltage measured at 3s does meet the threshold test, then the model proceeds to step S514. If the determination at step S506 is that the voltage measured at 3s does not meet the threshold test, then the model proceeds to step S510. If the determination at step S508 is that the voltage measured at 3.5s does meet the threshold test, then the model proceeds to step S514. If the determination at step S508 is that the voltage measured at 3.5s does not meet the threshold test, then the model proceeds to step S520. At step S510, an impedance measured at a frequency of 10e3 Hz is compared with a threshold (for example, the model may assess whether the impedance measured at a frequency of 10e3 Hz is less than or equal to 0). If the value meets the threshold test (e.g., if the impedance measured at a frequency of 10e3 Hz is less than or equal to 0) then the model proceeds to step S518. If the value does not meet the threshold test (e.g., if the impedance measured at a frequency of 10e3 Hz is greater than 0) then the model proceeds to step S514. The model also makes similar threshold of impedance measurements at steps S514, S516, and S518. At step S514, an impedance measured at a frequency of 20 Hz is compared with a threshold (for example, the model may assess whether the impedance measured at a frequency of 20 Hz is less than or equal to 0). If the value meets the threshold test (e.g., if the impedance measured at a frequency of 20 Hz is less than or equal to 0) then the model proceeds to step S518. If the value does not meet the threshold test (e.g., if the impedance measured at a frequency of 20 Hz is greater than 0) then the model proceeds to step S520. At step S516, an impedance measured at a frequency of 10 Hz is compared with a threshold (for example, the model may assess whether the impedance measured at a frequency of 10 Hz is less than or equal to 0). If the value meets the threshold test (e.g., if the impedance measured at a frequency of 10 Hz is less than or equal to 0) then the model proceeds to step S520. If the value does not meet the threshold test (e.g., if the impedance measured at a frequency of 10 Hz is greater than 0) then the model proceeds to step S514. At step S518, an impedance measured at a frequency of 10e3 Hz is compared with a threshold (for example, the model may assess whether the impedance measured at a frequency of 10e3 Hz is less than or equal to 0). If the value meets the threshold test (e.g., if the impedance measured at a frequency of 10e3 Hz is less than or equal to 0) then the model proceeds to step S522. If the value does not meet the threshold test (e.g., if the impedance measured at a frequency of 10e3 Hz is greater than 0) then the model proceeds to step S520. Step S518 is based on the same data as step S510. If, by following the decision tree 500, the model arrives at step S520, then the trained classification model indicates that the microbe present in the well is other (i.e., not E.coli). If, by following the decision tree 500, the model arrives at step S522, then the trained classification model indicates that the microbe present in the well is E.coli. This classification may be presented to a user in any suitable manner (e.g., through a display device as described above). The model then passes to step S524, where the decision tree 500 ends. During training, decision trees like that shown in Figure 5 are iteratively refined, for example using the XGBoost algorithm as described above. Figure 6 shows a confusion matrix for a trained classification model for Approach 2, and figure 7 shows the receiver operating characteristic (ROC) curves for the classifications, demonstrating the accuracy of Approach 2. Display of results Optionally, the results of the evaluation process may be displayed to a user. This allows a clinician to seethe results of the evaluation and take appropriate action. Once a result has been obtained for each well (or another end condition has been met) the test is terminated. The data is saved and the result is passed on to the display 1210, i.e. a GUI. The GUI displays the result(s), which may consist of the following items: Exception handle: error message; No growth detected: message informing the operator that bacterial growth has not been detected in the positive control (sample) well at the minimum required concentration, such that identification or classification may be inaccurate or not possible; Growth detected: message informing the operator that bacterial growth has been detected in the positive control (sample) well at the minimum required concentration, such that identification or classification is possible; and Microbe identification: a message informing the operator of the identification or classification of the microbe which is present or dominant in the sample (or, in a well or a subset of wells). All outcomes may be presented on the GUI while the test is running. Once the test is concluded, in addition to the GUI, a file is saved and stored on the instrument for future reference. As the measurement process is dynamic, an outcome may occur at any stage within the overall test time. Results will be displayed to the operator on the GUI as soon as they are available. The various methods described above may be implemented by a computer program. The computer program may include computer code arranged to instruct a computer to perform the functions of one or more of the various methods described above. The computer program and / or the code for performing such methods may be provided to an apparatus, such as a computer, on one or more computer readable media or, more generally, a computer program product. The computer readable media may be transitory or non-transitory. The one or more computer readable media could be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, or a propagation medium for data transmission, for example for downloading the code over the Internet. Alternatively, the one or more computer readable media could take the form of one or more physical computer readable media such as semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disc, and an optical disk, such as a CD-ROM, CD-R / W or DVD. In an implementation, the modules, components and other features described herein can be implemented as discrete components or integrated in the functionality of hardware components such as ASICS, FPGAs, DSPs or similar devices. In addition, the modules and components can be implemented as firmware or functional circuitry within hardware devices. Further, the modules and components can be implemented in any combination of hardware devices and software components, or only in software (e.g., code stored or otherwise embodied in a machine-readable medium or in a transmission medium). Unless specifically stated otherwise, as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as "receiving", "determining", "comparing ", "enabling", "maintaining," "identifying," or the like, refer to the actions and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices. The approaches described herein may be embodied on a computer-readable medium, which may be a non-transitory computer-readable medium. The computer-readable medium carrying computer-readable instructions arranged for execution upon a processor so as to make the processor carry out any or all of the methods described herein. The term "computer-readable medium" as used herein refers to any medium that stores data and / or instructions for causing a processor to operate in a specific manner. Such storage medium may comprise non-volatile media and / or volatile media. Non-volatile media may include, for example, optical or magnetic disks. Volatile media may include dynamic memory. Exemplary forms of storage medium include, a floppy disk, a flexible disk, a hard disk, a solid state drive, a magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with one or more patterns of holes, a RAM, a PROM, an EPROM, a FLASH-EPROM, NVRAM, and any other memory chip or cartridge. It is to be understood that the above description is intended to be illustrative, and not restrictive. Many other implementations will be apparent to those of skill in the art upon reading and understanding the above description. Although the present disclosure has been described with reference to specific example implementations, it will be recognized that the disclosure is not limited to the implementations described, but can be practiced with modification and alteration within the spirit and scope of the appended claims. Accordingly, the specification and drawings are to be regarded in an illustrative sense rather than a restrictive sense. The scope of the disclosure 5 should, therefore, be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. A computer-implemented method suitable for identifying or classifying microbes, the method for use with a system comprising a cartridge for receiving a sample, the cartridge comprising an array of wells, each well comprising an electrochemical transducer; wherein the method comprises:interrogating each of the electrochemical transducers, wherein interrogation of the electrochemical transducers comprises, for each well, generating a plurality of data series by repeatedly measuring a respective plurality of dependent variables; anddetermining, based on the plurality of data series from each well, an identification or classification of a microbe in each well.

2. The computer-implemented method of claim 1, wherein the determining is performed using a trained classification model.

3. The computer-implemented method of claim 2, wherein the trained classification model is trained using a gradient boosting algorithm.

4. The computer-implemented method of claim 2 or 3, wherein the plurality of dependent variables are individually weighted for the trained classification model.

5. The computer-implemented method of any preceding claim, wherein interrogation of the electrochemical transducers comprises, for each well, generating the plurality of data series by repeatedly measuring the respective plurality of dependent variables while controlling at least one independent variable.

6. The computer-implemented method of claim 5, wherein the at least one independent variable is selected from voltage, frequency, and time.

7. The computer implemented method of claim 5 or 6, wherein the plurality of dependent variables are selected from current, capacitance, impedance, a first derivative of the impedance, a second derivative of the impedance, phase, and voltage.

8. The computer-implemented method of any one of claims 5 to 7, wherein the plurality of dependent variables further comprises the concentration of microbial matter in the sample over time.

9. The computer-implemented method of claim 8, wherein the concentration of microbial matter is determined according to a numerical model.

10. The computer-implemented method of any preceding claim, wherein determining an identification or classification of a microbe in each well comprises identifying an organism species.

11. The computer-implemented method of any preceding claim, wherein determining an identification or classification of a microbe in each well comprises differentiating between different organism species.

12. A system for identifying microbial growth, the system comprising:a cartridge for receiving a sample that may comprise a microbe, the cartridge comprising an array of wells, each well comprising an electrochemical transducer; anda processor configured to perform a method according to any one of claims 1 to 10.

13. The system of claim 12, wherein the array of wells comprises a control subset of wells and a first subset of wells containing a first microbial sample.

14. The system of claim 13, wherein the array of wells further comprises a second subset of wells containing a second microbial sample.

15. The system of claim 13 or 14, wherein the array of wells further comprises a third subset of wells comprising an antimicrobial agent.

16. A computer readable medium comprising computer-executable instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1 to 11.s

Images