Classification of liquid handling operations using neural networks.

Abstract

To provide a method for classifying liquid handling procedures, a computer program and a computer-readable medium as well as a laboratory automation system, and simplify a configuration and / or control of the laboratory automation system.SOLUTION: A method for classifying liquid handling procedures includes: receiving measurement data encoding a measurement curve of measurements over time during at least a part of a liquid handling procedure; inputting the measurement data into a neural network; and calculating at least one quality value for the liquid handling procedure with the neural network.SELECTED DRAWING: None

Description

[Technical field]

[0001] The present invention relates to a method, a computer program and a computer readable medium for classifying liquid handling operations, and a laboratory automation system. [Background technology]

[0002] For example, laboratory automation systems are used to automate the work of laboratory technicians who test patients for specific diseases. Typically, samples of blood, urine, stool, etc. are taken from the patient and analyzed by biochemical techniques. Such techniques consist of various operations such as the addition of substances, incubation, and separation, and a measurement process that quantitatively or qualitatively measures the amount or presence of a substance indicative of a specific disease.

[0003] An important part of a laboratory automation system is the liquid handling system, which usually includes one or more pipettes, which are movable in three dimensions and can automatically aspirate liquid from a cavity and automatically dispense liquid into another cavity. The automatic transfer of liquid between different cavities using pipettes is sometimes called a pipetting procedure. The liquid handling system may also include one or more dispensing cannulas, which may be part of a dispenser module placed on the work surface of the laboratory system, or which may be movable in three dimensions and can automatically dispense liquid from a reservoir into a cavity. The term "dispensing" is understood as either dispensing liquid, such as a reagent or a diluent, from a reservoir fluidly connected to a dispensing tip, or aspirating liquid from a cavity and dispensing it from a pipette tip into another cavity.

[0004] Process safety is becoming more and more important in automated liquid handling systems. For this reason, new air pipettors typically measure the air pressure during liquid transfer. Classifying such signals is usually not easy, since there are multiple influencing parameters that affect the pipetting operation, such as liquid parameters, dynamic parameters, system and pipette tip parameters, and environmental parameters.

[0005] There are several ways in which the measurement data from the pressure sensor can be evaluated.

[0006] For example, an ideal pressure curve can be determined based on many correct pipetting operations. Pipetting operations can be evaluated by comparing the measured pressure curve with the ideal pressure curve. This pressure curve deviates from the ideal pressure curve by only a certain percentage in the case of correct pipetting operations. However, the ideal pressure curve may have to be established for each combination of volume, sample and tip type.

[0007] As another example, a theoretical pressure curve can be determined based on pipetting, sample and environmental parameters. After a pipetting operation, the theoretical pressure curve can be fitted to the measured pipetting curve. The fitting parameters can then be evaluated. Based on the evaluation, the liquid movement (i.e., the pipetting operation) may be evaluated as correct or incorrect. However, in this case, a theoretical model of the entire liquid handling system may have to be determined. Furthermore, the model may have to be adapted and validated for each combination of tip type, sample type and liquid handling system.

[0008] EP 1 745 851 A1 describes a pipetting device suitable for sorting liquids, which is based on a comparison of simulated and actual curves.

[0009] In WO 2012 068 610 A1 the density of a liquid is inferred by using measured pressure as input to a trained neural network.

[0010] US 2004 / 034 479 A1 and EP 1 391 734 A2 relate to a sample dispensing device and an automatic analyzer which use a neural network to analyze the waveform of pressure fluctuations.

[0011] Unver et al., "A fuzzy quality control-decision support system for improving operational reliability of liquid transfer operations in laboratory automation", Expert Systems with Applications, vo. 36, no. 4, 14 November 2018, pages 8064-8070, uses fuzzy logic to perform quality control in liquid transfer operations. Mohamad et al., "Learning a Multi-branch Neuronal Network from Multiple Sources for Knowledge Adaption in remote Sensing Imagery", Remote Sensing, vo.10, no 12, 1890, 27 November 2018, pages 1-18, describes the evaluation of image data using a neuronal network, and shows a neuronal network with different branches for different source data. Summary of the Invention [Problem to be solved by the invention]

[0012] An object of the present invention is to simplify the configuration and / or control of laboratory automation systems. [Means for solving the problem]

[0013] This object is achieved by the subject matter of the independent claims. Further exemplary embodiments emerge from the cited claims and the following description.

[0014] A first aspect of the invention relates to a method for classifying liquid handling operations. The method can be performed automatically, for example by a controller of a laboratory automation system. The liquid handling operations can be pipetting operations, parts of pipetting operations and / or dispensing operations. The automatic transfer of liquids between different cavities with a pipette performed by a laboratory automation device can be referred to as a pipetting operation. The automatic dispensing of liquids with a pipette and / or a dispensing cannula performed by a laboratory automation device can be referred to as a dispensing operation. The dispensing operation can be a part of a pipetting operation.

[0015] According to one embodiment of the invention, the method includes receiving measurement data encoded with a measurement curve for measurements over time during a liquid processing operation. The measurement data may include sensor data related to the liquid processing, such as pressure and / or flow measurements of liquids and / or gases.

[0016] It is also possible to measure the filling level of pipette tips or containers. For example, the volume of liquid aspirated can be measured by volume sensing in the pipette tip. Also, by sensing the liquid level before and after aspirating or dispensing, the change in volume in the source or destination cavity can be measured and the volume difference can be calculated using the known cross-section of the cavity.

[0017] The measurement data can be obtained by sensors in the laboratory automation system, for example measuring the pressure and / or flow rate of gas and / or liquid in the lines to the pipettes of the laboratory automation system, or the measurements can be performed in the space connected to the pipette and / or dispensing cannula, where the pressure is changed to perform the liquid handling operation.

[0018] Pressure applied to the line can be applied by a pump and can control the aspiration and dispensing of liquid by the pipette tip. The applied pressure and the liquid flowing through the pipette can change the pressure in the line, which can change the flow in the line.

[0019] According to one embodiment of the invention, the method further comprises inputting the measurement data into an (artificial) neural network and calculating, using the neural network, at least one quality value for the liquid processing operation.

[0020] The neural network may be provided as a software module and / or software library, which is complemented by parameterization created from previously validated training data. The training data may be a set of measurement data and any liquid processing data of a number of different previously validated liquid processing operations. The liquid processing data may encode configurations and / or settings of the laboratory automation system. The training data may include one or more quality values ​​for each liquid processing operation.

[0021] The neural network may be a deep neural network, i.e., it may include multiple layers, such as convolutional layers and fully connected layers. The neural network may include at least two convolutional layers and / or at least two fully connected layers connected in series. Below, we define the different types of layers described herein.

[0022] A layer may be a set of neurons having inputs for receiving input values, e.g. from a previous layer, and outputs for sending output values, e.g. to a next layer. The neurons may have weights and a function for calculating an output value from the input values ​​depending on the weights. The layers may be provided with objects in an object-oriented programming language. The number, weights and functions of the neurons may be provided as parameter data, which may be, e.g., coded in the object. The parameter data may be considered as the parameterization of the neural network, whereas the layers and their interconnections may be considered as the structure of the neural network.

[0023] Measurement data from real measurements in a laboratory automation system may be input to a neural network, which is trained to validate the measurement data. The validation may be a classification, such as a classification into successful and unsuccessful liquid handling operations. The validation may also be one or more values ​​indicative of physical quantities, which are present in the respective liquid handling operations, such as aspirated and / or dispensed volumes. Furthermore, possible physical quantities may be physical properties of the pipetted and / or dispensed sample liquids, such as specific density, viscosity, surface tension and / or wettability of the pipette tip (and / or dispensing cannula) surface by the sample.

[0024] In general, neural network-based machine learning algorithms can be used to evaluate measurement data to verify and / or classify liquid movements in laboratory automation systems. The trained neural network can be used as an error detection assistant.

[0025] The neural network can even be tuned during and / or after the operation of the laboratory automation system. The error detection rate may be adjusted against the false positive rate, and the training set may be supplemented with validated and / or classified measurement curves. Customers using the laboratory automation system can validate the measurement curves and improve the predictive ability of the neural network.

[0026] According to one embodiment of the invention, the method further comprises controlling a laboratory automation system using the quality value. The method can be implemented by a controller of the laboratory automation system during operation of the laboratory automation system. For example, the quality value may classify a liquid handling operation as correct or incorrect, the quality value may indicate that the aspirated volume is too small, etc. If the liquid handling operation is deemed incorrect, the assays and / or samples processed by the laboratory automation system in the liquid handling operation may be marked as incorrect and / or discarded.

[0027] According to one embodiment of the invention, the method further comprises marking the liquid processing operation as erroneous if the quality value indicates a failed liquid processing operation and / or discarding the results of the liquid processing operation if the quality value indicates a failed liquid processing operation.

[0028] According to one embodiment of the invention, the method further comprises repeating the assay and / or sample processing with the liquid handling operation if the quality value indicates a failed liquid handling operation.

[0029] According to one embodiment of the invention, the measurement data comprises a vector of measurements arranged in a time series. Pressure, flow rate and / or more general liquid processing sensor data can be measured over time, i.e. a measurement curve can be determined. The measurement data may comprise time-discrete measurements of this curve. The measurements may be concatenated into a time series vector. Such a vector can be considered as a one-dimensional digitized image of a continuous physical quantity and may therefore be particularly suitable to be input into a neural network.

[0030] According to one embodiment of the present invention, the method further comprises inputting liquid handling data into the neural network, the liquid handling data encoding the configuration and / or settings of the laboratory automation system performing the liquid handling operations. Besides the measurement data, liquid handling data, i.e. data encoding the properties of the laboratory automation system, such as the pipette tips actually used, can also be input into the neural network. In other words, the neural network may be trained for different configurations and / or settings of the laboratory automation system. In this way, the same neural network can be used in different application scenarios.

[0031] The liquid handling data may include information about the liquid, such as density, viscosity and / or type of liquid. The liquid handling data may include information about the pipette and / or disposable pipette tips (size, type, maximum capacity, etc.). The liquid handling data may include information about liquid handling operations and / or control parameters, such as aspirating and / or dispensing speed, amount of liquid to aspirate and / or dispense, length of aspirating and / or dispensing, control parameters of pumps, etc.

[0032] According to one embodiment of the present invention, the measurement data and the liquid processing data are concatenated into a vector before being input to the neural network. The measurement data and liquid handling data are input to one input layer of the neural network. The measurement data and liquid handling data can be treated as a single piece of input information.

[0033] According to one embodiment of the present invention, the measurement data and the liquid processing data are input to different input layers of the neural network. The two types of data may also be input to different input layers, which may be connected to different other layers of the neural network. In this way, the two different data sets may be processed differently by different configurations of layers before being processed by the same layer.

[0034] According to one embodiment of the invention, the neural network includes a measurement data branch organized in at least one layer and a liquid processing data branch organized in at least one layer, where the measurement data may be input to an input layer of the measurement data branch and the liquid processing data may be input to an input layer of the liquid processing data branch.

[0035] The measurement data branch may include at least two convolutional layers. However, it may be possible for the measurement data branch to have only one input layer. The liquid processing data branch may have only one input layer, but may also include one, two or more fully connected layers.

[0036] According to one embodiment of the present invention, the neural network includes a branch of a fully connected layer. For example, the output of the measurement data branch and the output of the liquid processing data branch are input to the branch of the fully connected layer. The branch of the fully connected layer may include at least two fully connected layers. The branch of the fully connected layer may include a probability layer at its end, which creates probability values ​​for different classification types.

[0037] In general, the neural network, especially the branches of the fully connected layer, can output a number of quality values. Such quality values ​​may include classification values, which may be percentage values ​​indicating the probability of a particular classification type. Such quality values ​​may also be estimates that estimate certain physical quantities that may occur during liquid processing operations.

[0038] According to one embodiment of the present invention, the neural network outputs a classification value that classifies the liquid handling operation. For example, the classification value may indicate at least one of: correct operation, clogging, aspiration of air, lack of sample, air bubbles, foaming, blocked tip, leaking, etc. All or some types of problems that may occur during a liquid handling operation may be classified. Also, the classification may be limited to only indicating whether the operation was performed correctly or not.

[0039] According to one embodiment of the present invention, the neural network outputs estimates that estimate physical quantities of a liquid handling operation. For example, the estimates estimate at least one of a dispense volume and an aspiration volume. The neural network may be trained as a model of the laboratory automation system. It outputs physical quantities, which are determined from measurement data. It is not necessary to determine the model based on mathematical functions in which physical relationships are encoded.

[0040] According to one embodiment of the invention, a liquid handling operation includes lowering a pipette into a cavity containing liquid. During the lowering of the pipette into the liquid, i.e. before the pump begins operation, pressure, flow and / or liquid handling sensor data can be measured.

[0041] According to one embodiment of the present invention, the liquid handling operation comprises reducing the pressure in the pipette to aspirate liquid into the pipette and / or increasing the pressure in the pipette to dispense liquid into the pipette. The pressure can be reduced and / or increased by a pump connected to the pipette. The measurement data may comprise measurements upon aspirating liquid into the pipette and / or the measurement data may comprise measurements upon dispensing liquid from the pipette. It is noted that the validation by the neural network can be performed only during aspiration, only during dispensing and / or both.

[0042] Another aspect of the invention relates to a computer program for classifying liquid handling operations, which, when executed by a processor, is adapted to carry out the steps of the method described above and below. The computer program can be executed on a controller of a laboratory automation system and / or on a computing device, such as a PC, communicatively interconnected with the laboratory automation system. The method can also be implemented by an embedded microcontroller.

[0043] Another aspect of the invention relates to a computer readable medium on which such a computer program is stored. The computer readable medium may be a floppy disk, a hard disk, a Universal Serial Bus (USB) storage device, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM) or a flash memory. The computer readable medium may also be a data communication network such as the Internet from which the program code can be downloaded. In general, the computer readable medium may be a non-transitory or a transitory medium.

[0044] Yet another aspect of the present invention relates to a laboratory automation system.

[0045] According to one embodiment of the present invention, the laboratory automation system comprises a liquid handling arm for carrying a pipette and / or a dispensing cannula, a pump for varying the pressure in a space connected to the pipette in order to aspirate and dispense liquid in the pipette, a sensor device for performing measurements in the space connected to the pipette, and a control device for controlling the pump and receiving measurement data from the sensor device.

[0046] The measurements may be any measurements related to a fluid processing operation, such as pressure, flow rate, volume change measurements, etc. The measurement data may be or include fluid processing sensor data.

[0047] Furthermore, the controller may be adapted to carry out the methods as described above and below.The controller may store a neural network as described above and below.

[0048] It is to be understood that features of the methods as described above and below may also be features of the control devices, computer programs and computer readable media as described above and below, and vice versa.

[0049] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. [Brief description of the drawings]

[0050] Hereinafter, embodiments of the present invention will be described in more detail with reference to the accompanying drawings.

[0051] [Figure 1] FIG. 1 illustrates a schematic diagram of a laboratory automation system according to one embodiment of the present invention. [Diagram 2] FIG. 2 shows a flow diagram of a method for classifying liquid processing operations according to one embodiment of the present invention. [Diagram 3] FIG. 3 shows the measurement curve during the pipetting operation. [Figure 4] FIG. 4 shows several measurement curves when there is a blockage during the pipetting operation. [Diagram 5] FIG. 5 shows several measurement curves when air bubbles are present during the pipetting procedure. [Figure 6] FIG. 6 shows a schematic of the input vectors of a neural network used in one embodiment of the present invention. [Figure 7]FIG. 7 shows a schematic layout of a neural network used in one embodiment of the present invention. [Figure 8] FIG. 8 shows a schematic diagram of yet another layout of a neural network for use in one embodiment of the present invention. [Figure 9] FIG. 9 illustrates, in schematic form, yet another layout of a neural network for use in one embodiment of the present invention.

[0052] Reference symbols used in the drawings and their meanings are listed below. Basically, the same reference symbols are used for the same parts in multiple drawings. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0053] A laboratory automation system 10 is shown generally in Figure 1. The system includes an automatically movable pipette arm 12 having a pipette 14 attached thereto. The pipette 14 may include a disposable pipette tip 15, which may also be grasped and discarded by the pipette arm 12. As shown in Figure 1, the pipette tip 15 is lowered into a container 16, which may be a well of a multi-well plate, a test tube containing a sample, a container containing a reagent, or the like.

[0054] The pipette arm 12 can move the pipette 14 and pipette tip 15 in three dimensions and can lower the pipette tip 15 into and raise the pipette tip from a container 16. The container 16 and possibly the pipette tip 15 may contain a liquid 18, such as a sample or a reagent. The pipette 14 and its tip 15 are used to transfer and / or transport volumes of liquid between different containers 16.

[0055] Laboratory automation system 10 may include dispensing cannulas. These cannulas may be connected to pipette arm 12, which may also be considered a liquid handling arm. The dispensing cannulas may be connected to a reservoir and may be used to dispense liquid into container 16.

[0056] Additionally, the laboratory automation system 10 includes a pump 20 connected to the pipette 14 via a hose 22. The pump 20 can be used to apply pressure to the hose 22 and the pipette 14, causing the pipette 14 to aspirate or dispense liquid 18.

[0057] A sensor device 24, which may be attached to the hose 22 and / or the pipette 14, is adapted to measure the pressure and / or flow rate within the hose 22 and / or the pipette 14. Measurement data obtained by the sensor device 24 may be used to verify pipetting operations as described herein.

[0058] The controller 26 of the laboratory automation system 10 may be part of or connected to the laboratory automation system 10 and may control the pipette arm 12, the pump 20, and may receive measurement data from the sensor device 24.

[0059] A flow diagram of a method for classifying liquid processing operations that may be performed by controller 26 is shown in FIG.

[0060] In step S10, measurement data 56 is generated. Typically, the sensor device 24 of the laboratory automation system 10 can measure pressure and / or flow over time, from which a measurement curve can be formed. During measurement, measurements may be taken over time and transmitted to the controller 26.

[0061] Measurement curves 28, in particular pressure curves 28, during a pipetting operation are shown in Fig. 3. It should be noted that everything that is said below regarding pressure curves and pressure measurements also applies to flow curves and flow measurements, and everything that is said below may also be relevant, if applicable, for dispensing operations carried out with a dispensing cannula.

[0062] The generation of the measurement data 56 in step S10 can be performed in parallel with the control of the laboratory automation system 10, which can also be performed by the control device .

[0063] Typically, the liquid 18 is moved between the two cavities 16 by reducing the pressure in the pipette 14 and aspirating the liquid 18 into the pipette 14, which can be achieved by appropriately controlling the pump 20. The pressure in the pipette 14 is then increased, which can be achieved by appropriately controlling the pump 20 to dispense the liquid 18 from the pipette 14. Note that the pipette arm 12 may move the pipette 14 between the first and second cavities 16 before, during or after aspirating and dispensing.

[0064] 3, pressure curve 28 shows the actual pressure measured from gripping pipette tip 15 to finally lowering pipette tip 14. Here, time flows from left to right. In particular, aspiration curve 30 and dispensing curve 32 are shown enlarged.

[0065] First, the pipette tip 15 is grasped (34) and a movement (36) of the pipette arm 12 to the first cavity 16 takes place. At (38) aspiration is started. It can be seen that a drop in pressure creates an underpressure for aspirating liquid 18 into the pipette tip 15. At (40) the aspiration ends and the pressure returns to the mean value. Then a further movement (42) of the pipette arm 12 takes place to the second cavity 16. At (44) the dispensing of liquid 18 starts and at (46) the dispensing ends. As can be seen, the pressure is now increased so that liquid 18 is dispensed by an overpressure in the pipette tip 15. Finally, a movement (48) of the pipette arm 12 takes place to a waste container where the pipette tip 15 is dropped (50).

[0066] When aspirating (30) and dispensing (32) are performed correctly, the pressure curve will look like that in Figure 3. However, errors during aspirating and / or dispensing (32) will result in a pressure curve with a different shape.

[0067] As an example, aspiration curves 30 when the pipetting operation is not performed correctly are shown in Figures 4 and 5. In particular, Figure 4 shows a number of aspiration curves 30 when a clog is present during the pipetting operation. It can be seen that all of these curves deviate in some similar way from the optimal curve as shown in Figure 3. Figure 5 shows a number of aspiration curves 30 when air bubbles are present during the pipetting operation. Again, it can be seen that all of these curves deviate in the same way from the optimal curve as shown in Figure 3.

[0068] Returning to FIG. 2, in step S12, measurement data 56 is received by the controller 26, which encodes the measurement curves 28, 30 and / or 32 for measurements over time during the performance of a pipetting operation.

[0069] The controller 26 can create a data structure in the form of a vector from the measured values, where the measured values ​​are arranged in chronological order, and the data vector can be supplemented with other data, such as configuration parameters and / or parameter settings of the laboratory automation system 10.

[0070] A data vector 52 that may be generated by the controller 26 is shown generally in Figure 6. The data vector 52 is made up of fluid process data 54 and measurement data 56.

[0071] The liquid handling data 54 may consist of configuration parameters and / or settings 55, which may depend, for example, on the type of liquid handling operation actually performed, the type of pipette tip 15 used, the type of liquid 18, etc. The controller 26 may assemble such data and place it into a data vector 52.

[0072] It should be noted that the entries and / or values ​​55 in the fluid processing data 54 may be of different sizes and / or different formats.

[0073] The measurement data 56 may consist of measurements 53. The measurement data 56 may be pre-processed by the controller 26, for example, to fit into a data vector 52 of a particular length.

[0074] The measurement data 56 may be arranged as a vector 52 of chronologically ordered measurements 53, i.e. the higher the exponent of a measurement 53 the earlier the time that the measurement was taken.

[0075] Returning to Figure 2, in step S14, the measurement data 56 and optionally the liquid processing data 54 are input into a neural network which is used to calculate at least one quality value 74, 86 for the liquid processing operation. An example of a neural network is described in relation to Figures 7 to 9.

[0076] For example, the layout and / or structure of a neural network 57 used in one embodiment of the present invention is shown generally in Figure 7. In Figure 7, the layout is a sequence of connected layers 58-72, each made up of a set of neurons.

[0077] It should be noted that besides its layout, neural network 57 also includes configurations / parameterizations for its layers, such as the number of neurons and / or the number of inputs and / or outputs for each neuron, etc. Additionally, each layer also includes weights for its inputs and functions for its outputs, and calculates the respective output values ​​based on the weights.

[0078] The weights may be determined during training of the neural network 57, which is provided with a number of already validated training data sets, such as the curves 28, 30, 32 as shown in Figures 3 to 5, optionally together with the corresponding liquid processing data. This may be achieved, for example, by backpropagation.

[0079] In the example of Figure 7, the measurement data 56 and the liquid processing data 54 may be concatenated, for example as one vector 52 as shown in Figure 6, before being input to the neural network 57. However, it is also possible to input only the vector of measurement data 56 to the neural network 57. This may be the case if the neural network has only been trained for one configuration and / or type of liquid processing operation.

[0080] 7 includes (in that order) an input layer 58, a reshape layer 60, multiple pairs of convolutional layers 62 and pooling layers 64, a dropout layer 66, a smoothing layer 68, a number of fully connected layers 70, and a probability layer 72. Further dropout layers 66 may be disposed between the fully connected layers 70.

[0081] The input layer 58 receives vectors of input data (such as measurement data 56 and / or fluid process data 54) and / or pre-processes the input data in a first step.

[0082] The reshape layer 60 changes the dimensions of the output data relative to the input data. If the correct format is provided directly, the reshape layer 60 may not be necessary.

[0083] The convolutional layer 62 applies one or more filters (also called kernels) to the local receptive fields of the input data to create feature maps. In this example, the output of the reshape layer and / or the local receptive fields may be one-dimensional.

[0084] The pooling layer 64 is a layer that condenses the information from the previous layer, for example by taking the maximum value of a region.

[0085] The dropout layers 66 remove some of the feedback information during training in order to generalize the learning. One, several, or all of the dropout layers can be optional to increase the generality of the predictions.

[0086] The planarization layer 68 converts the multi-dimensional data into one-dimensional data.

[0087] A fully connected layer 70 is a layer that has full one-to-one connections with the previous layer.

[0088] A probability layer is a layer that determines the probability value of a classifier from unnormalized inputs. A probability layer may be a fully connected layer with outputs normalized by a softmax activation function to sum to one.

[0089] Finally, the neural network 57 outputs one or more classification values ​​74, which are probability values ​​that classify the liquid handling operation. For example, the classification values ​​74 may indicate correct operation, clog, aspirate, short sample, air bubble, foam, blocked tip, or leak.

[0090] Another layout of a neural network 57 that may be used in an embodiment of the invention is shown generally in Figure 8. In Figure 8, the measurement data 56 and the fluid process data 54 are input to different input layers 58, 82 of the neural network 57. Furthermore, the neural network 57 includes a measurement data branch 76 and a fluid process data branch 78.

[0091] The measurement data 56 is input to an input layer 58 of the measurement data branch 76 and the fluid process data 54 is input to an input layer 82 of the fluid process data branch 78 .

[0092] The measured data branch 76 may include at least two pairs of convolutional layers 62 and pooling layers 64. Additionally, the measured data branch 76 may include a reshape layer 60 after the input layer 58, and finally a flattening layer 68.

[0093] The liquid treatment data branch 78 is made up of only an input layer 82 to which the liquid treatment data 54 is input.

[0094] The outputs of the measurement data branch 76 and the fluid process data branch 78 are input to a concatenation layer 84 that concatenates the outputs, such as two vectors, together along a particular dimension.

[0095] The output of the concatenation layer 84 is input to a fully connected layer branch 80 of the neural network 57. The fully connected layer branch 80 may include at least two fully connected layers 70. A dropout layer 66 may be disposed between the fully connected layers 70. The output of the fully connected layer branch 80 is input to a probability layer 72, such as that shown in FIG. 7.

[0096] Neural network 57 in FIG. 7 also has a measurement data branch 76, or a more general convolutional branch, and a fully connected layer branch, which may be assumed to be connected in series.

[0097] Yet another layout of a neural network 57 that may be used in an embodiment of the present invention is shown generally in FIG.

[0098] 8, the neural network 57 of Figure 9 includes a fully connected layer branch 80 which receives the outputs of the measurement data branch 76 and the fluid treatment data branch 78. However, the measurement data branch 76 and the fluid treatment data branch 78 each consist of only one input layer 58, 82.

[0099] In contrast to the neural network 57 of Figures 7 and 8, the neural network 57 of Figure 9 has been trained to output one or more estimates 86 that estimate physical quantities of a liquid handling operation. For example, the estimates 86 may be the amount dispensed or the amount aspirated. In this case, the training data needs to be provided with the corresponding estimates.

[0100] It may be possible for the neural network 57 of Figures 7 and 8 to output an estimate 86 like the neural network 57 of Figure 9, and for the neural network 57 of Figure 9 to output a probability value 74 like the neural network 57 of Figures 7 and 8.

[0101] Returning to FIG. 2, in step S16, the assays and / or samples processed by the laboratory automation system 10 may be marked with one or more quality values ​​74, 86 determined by the neural network 57.

[0102] This may also be to discard the assays and / or samples that were processed in an incorrectly performed liquid handling operation as indicated by the quality values ​​74, 86 and / or to process this assay and / or sample a second time.

[0103] Although the present invention has been illustrated and described in detail in the drawings and the above description, such illustrations and descriptions are not to be considered as limiting or illustrative, and the present invention is not limited to the disclosed embodiments. Those skilled in the art can understand and effect other variations to the disclosed embodiments by practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or controller or other unit may fulfill the functions of several items recited in a claim. The mere fact that certain measures are recited in mutually different recited claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting their scope. [Explanation of symbols]

[0104] 10 Laboratory Automation Systems 12 Pipette Arm 14 Pipettes 15 Pipette Tips 16 Container 18 liquid 20 Pump 22 Hose 24 Sensor device 26 Control device 28 Pressure Curve 30 suction curve 32 Dispensing curve 34 Pipette tip grip 36 Pipette Arm Movement 38 Start suction 40 Suction completed 42 Pipette arm movement 44 Start dispensing 46 Dispensing completed 48 Pipette Arm Movement 50 Pipette Tip Drops 52 Data Vectors 53 Measurements 54 Liquid Processing Data 55 Configuration Parameters 56 Measurement Data 57 Neural Networks 58 Input Layer 60 Reshape Layers 62 Convolutional Layer 64 pooling layers 66 Dropout Layer 68 Planarization layer 70 fully connected layer 72 Probabilistic Decision Layer 74 Quality Value, Classification Value 76 Measurement Data Branch 78 Liquid Processing Data Branch 80 Branching of fully connected layers 82 Second Input Layer 84 Connected layer 86 Quality Value, Estimated Value

Claims

1. 1. A method for grouping a plurality of liquid processing operations, comprising: receiving measurement data (56) in which a measurement curve (28, 30, 32) of a measurement over time during a liquid processing operation is encoded; inputting said measurement data (56) into a neural network (57); inputting liquid processing data (54) into said neural network (57); calculating, using said neural network (57), at least one quality value (74, 86) for said liquid treatment operation; the liquid handling data (54) encoding configurations and / or settings of a laboratory automation system (10) that performs the liquid handling operations; The liquid processing operation is classified as correct or incorrect according to the quality value; The method comprises: said measurement data (56) and said liquid processing data (54) are concatenated into a vector (52) before being input into said neural network (57); 4. The method of claim 3, wherein said measurement data (56) and said liquid processing data (54) are input to one input layer of said neural network (57).

2. The method of claim 1 , wherein the measurement data (56) comprises a vector (52) of time-sequenced measurements (53).

3. the neural network (57) outputs a classification value (74) that classifies the liquid processing operation; The method according to any one of claims 1 to 2, wherein the classification value (74) indicates at least one of the following: correct operation, clogging, aspiration of air, lack of sample, air bubbles, foaming, blocked tip, leakage.

4. The method according to any one of claims 1 to 3, wherein the neural network (57) outputs an estimate (86) for estimating a physical quantity of the liquid processing operation, the estimate (86) estimating at least one of a dispense volume, an aspiration volume.

5. The liquid treatment operation comprises: Aspirating (30) liquid (18) into the pipette (14) by reducing the pressure in said pipette (14); The method of any one of claims 1 to 4, comprising at least one of dispensing (32) liquid in the pipette (14) and / or dispensing cannula by increasing pressure in the pipette (14) and / or dispensing cannula.

6. A computer program for classifying liquid processing operations, the program being adapted to carry out the steps of the method according to any one of claims 1 to 5 when executed by a processor.

7. A computer readable medium having stored thereon the computer program of claim 6.

8. a liquid handling arm (12) for carrying pipettes (14) and / or dispensing cannulas; a pump (20) for varying the pressure in a space (22) connected to said pipette (14) in order to aspirate and dispense a liquid (18) in said pipette (14); a sensor device (24) connected to said pipette (14) for performing measurements in said space (22); a control device (26) for controlling the pump (20) and for receiving measurement data (56) from the sensor device (24); A laboratory automation system (10), wherein the control device (26) is adapted to perform the method according to any one of claims 1 to 5.

Images