Adaptive sorting for particle analyzers

Abstract

A cell sorting system that automatically generates a sorting strategy based on an example of a target event provided by an operator. The target event may be selected using measurements ranging from conventional flow cytometry measurements to derived measurements (which are costly in calculation for complex measurements such as images).

Description

[0001] This application is a divisional application, and its original application is a PCT application with application number PCT / US2019 / 055225 and application date October 8, 2019, and entered the Chinese national phase on April 23, 2021, with application number 201980070447.7 and name “Adaptive Sorting of Particle Analyzers”. Technical Field

[0002] The present invention relates generally to the field of automated particle assessment technology and, more particularly, to sample analysis and particle characterization methods. Background Art

[0003] Particle analyzers (e.g., flow cytometers and scanning cytometers) are analytical tools that characterize particles based on electro-optical measurements such as light scattering and fluorescence. In a flow cytometer, for example, particles in a fluid suspension (e.g., molecules, microbeads bound to an analyte, or individual cells) pass through a detection region where they are exposed to excitation light, typically from one or more lasers, and the light scattering and fluorescence properties of the particles are measured. Particles or their components are typically labeled with fluorescent dyes for detection. Various different particles or components can be detected simultaneously by labeling them with fluorescent dyes with different spectral characteristics. In some embodiments, the analyzer includes multiple photodetectors, one for each scattering parameter to be measured and one or more for each different dye to be detected. For example, some embodiments include a spectral structure in which more than one sensor or detector is used for each dye. The data obtained includes the signals measured for each of the light scattering detectors and the fluorescence emission.

[0004] The particle analyzer may further include means for recording the measured data and analyzing the data. For example, a computer connected to the detection electronics may be used for data storage and analysis. For example, the data may be stored in a tabular format, where each row corresponds to data for one particle and the columns correspond to each measured feature. Storing data from a particle analyzer using a standard file format (e.g., the "FCS" file format) facilitates analysis of the data using a separate program and / or machine. When current analysis methods are used, the data is typically displayed in the form of a one-dimensional histogram or a two-dimensional (2D) graph for ease of visualization, but other methods may be used to visualize multi-dimensional data.

[0005] For example, parameters measured using flow cytometry typically include light scattered by the particles at a narrow angle at the excitation wavelength primarily in the forward direction (referred to as forward scatter (FSC)); the excitation light scattered by the particles in a direction orthogonal to the excitation laser (referred to as side scatter (SSC)); and light emitted by fluorescent molecules in one or more detectors (used to measure signals over a certain spectral wavelength range), or light emitted by fluorescent dyes detected primarily in the specific detector or detector array. Different cell types can be identified by light scattering characteristics and fluorescence emissions obtained / generated by labeling various cellular proteins or other components with fluorescent dye-labeled antibodies or other fluorescent probes.

[0006] Flow cytometers and scanning cytometers are available from, for example, BD Biosciences (San Jose, CA). Flow cytometry is described in, for example, Landy et al. (eds.), Clinical Flow Cytometry, Annals of the New York Academy of Sciences, Vol. 677 (1993); Bauer et al. (eds.), Clinical Flow Cytometry: Principles and Applications, Williams & Wilkins (1993); Ormerod (ed.), Flow Cytometry: A Practical Approach, Oxford University Press (1994); Jaroszeski et al. (eds.), Flow Cytometry Protocols, Methods in Molecular Biology, Vol. 91, Humana Press (1997); and Shapiro, Practical Flow Cytometry, 4th Edition, Wiley-Liss (2003); all of which are incorporated herein by reference. Fluorescence imaging microscopy is described in, for example, Pawley (ed.), Handbook of Biological Confocal Microscopy, 2nd ed., Plenum Press (1989), which is incorporated herein by reference.

[0007] The data obtained by analyzing cells (or other particles) using multicolor flow cytometry is multidimensional, where each cell corresponds to a point in the multidimensional space defined by the measured parameters. Cell populations or particle populations are identified as clusters of points in the data space. Clusters and populations can be manually identified by drawing gates around a population displayed in one or more two-dimensional graphs of the data (called "scatter plots" or "dot plots"). Alternatively, clusters can be automatically identified and gates defining the boundaries of the population can be automatically determined. Examples of methods for automatic gating can be found in the following publications, for example, U.S. Patents No. 4,845,653; 5,627,040; 5,739,000; 5,795,727; 5,962,238; 6,014,904; 6,944,338; and U.S. Patent Publication No. 2012 / 0245889, each of which is incorporated herein by reference.

[0008] Flow cytometry is an effective method for analyzing and separating biological particles (e.g., cells and constituent molecules), and therefore, it is widely used in diagnostics and therapeutics. The method utilizes a fluid medium to linearly separate particles so that the particles can be aligned to pass through a detection device. Individual cells can be distinguished based on their location in the fluid medium and the presence or absence of a detectable marker. Therefore, flow cytometry can be used to characterize and generate diagnostic profiles of biological particle populations.

[0009] Separation of biological particles has been achieved by adding a sorting or collection function to flow cytometers. Particles in the separation stream that have been detected to have one or more desired characteristics are individually separated from the sample stream by mechanical or electrical separation. This flow sorting method has been used to sort different types of cells, separate sperm containing X and Y chromosomes for animal breeding, sort chromosomes for genetic analysis, and isolate specific organisms from complex biological populations.

[0010] Gating is used to help understand and classify the large amount of data that may be generated by a sample. Given the large amount of data presented by a given sample, there is a need to effectively control the graphical display of that data.

[0011] Fluorescence activated particle sorting or cell sorting is a specialized form of flow cytometry. It provides a method for sorting a heterogeneous mixture of particles into one or more containers, one cell at a time, based on the specific light scattering and fluorescence properties of each cell. It records the fluorescent signals from individual cells and physically separates the specific cells of interest. The abbreviation FACS is a trademark of and owned by Becton Dickinson and may be used to refer to equipment that performs fluorescence activated particle sorting or cell sorting.

[0012] The particle suspension is placed near the center of a narrow, fast-flowing stream. The stream is arranged so that, on average, there is a large separation between particles relative to their diameters as they randomly arrive (Poisson process) at the detection region. A vibration mechanism causes the outflowing fluid medium to steadily break up into single droplets containing the particles previously characterized in the detection region. The system is typically tuned so that the probability of more than one particle being present in a droplet is low. If a particle is classified as being collected, an electrical charge is applied to the flow cell and outflow stream over a period of time to form one or more droplets that detach from the stream. These charged droplets then pass through an electrostatic deflection system that transfers the droplets to a target container based on the charge applied to the droplets.

[0013] The sample may include thousands, if not millions, of cells. The cells may be sorted to purify the sample to cells of interest. The sorting process typically identifies three types of cells: cells of interest, non-cells of interest, and cells that cannot be identified. In order to sort cells with high purity (e.g., a high concentration of cells of interest), the cell sorter that generates the droplets typically electronically aborts the sorting if the desired cell is too close to another undesirable cell, thereby reducing contamination of the sorted population by inadvertently including undesirable particles in droplets containing particles of interest. Summary of the invention

[0014] In one innovative aspect, a computer-implemented method is provided. The method is performed under the control of one or more processing devices. The method includes receiving, with a particle analyzer, a reference sorting standard applicable to a sample from a communication device. The method includes identifying a candidate event classifier applicable to the sample based at least in part on the reference sorting standard and the sample. The method includes generating a sorting strategy, the sorting strategy including at least one event classifier from the candidate event classifiers. The method includes generating an indicator indicating the accuracy of the sorting strategy. The method includes determining that the indicator meets or exceeds a minimum accuracy threshold for the sample. The method includes configuring the particle analyzer to classify the particles of the sample based at least in part on multidimensional measurements of the particles and the sorting strategy.

[0015] In some embodiments, the reference sorting standard may include gate information identifying a range of measurements used to classify the particles. In some embodiments, the reference sorting standard may include an image showing reference particles to be collected from the sample.

[0016] The method may also include transmitting a sorter configuration to the particle analyzer, wherein the sorter configuration represents the sorting strategy; and adjusting a sorting circuit based at least in part on the sorter configuration. An example of the sorting circuit is a field programmable gate array.

[0017] In some embodiments, the multidimensional measurements received from the particle analyzer may include measurements of light emitted by the particles in the form of fluorescence. The light emitted by the particles in the form of fluorescence may include light emitted by antibodies bound to the particles in the form of fluorescence.

[0018] In some embodiments, the indicator may include an f-measure value of the sorting strategy. The precision component of the f-measure value represents the purity used to sort the sample according to the sorting strategy. The recall component of the f-measure value represents the yield used to sort the sample according to the sorting strategy.

[0019] Some embodiments of the method may include storing the candidate event classifier in a data storage device, wherein the candidate event classifier includes at least two of the following: (a) calculating a sorting classifier, which includes: a score and an associated target or non-target score; (b) comparing the score applicable to the particle measurement with the associated target or non-target score applicable to sorting; (c) a transformation that converts one or more measurements of a particle into an approximate measurement of the particle; (d) a parameter projection that converts a population represented by the reference sorting standard in a non-sortable first parameter space into a population in a sortable second parameter space; and (e) feature extraction that generates vector features using images or high-dimensional measurements, wherein the feature vector is used to classify the particle.

[0020] The method may include receiving a hardware identifier from the particle analyzer, the hardware identifier indicating sorting circuitry implemented in the particle analyzer; in such embodiments, identifying the candidate event classifier may further be based at least in part on the hardware identifier.

[0021] In another innovative aspect, a system is provided. The system includes one or more processing devices and a computer-readable storage medium (including instructions). When the instructions are executed by the one or more processing devices, the system receives a reference sorting standard applicable to a sample from a communication device using a particle analyzer; identifies a candidate event classifier applicable to the sample based at least in part on the reference sorting standard and the sample; generates a sorting strategy, the sorting strategy including at least one event classifier from the candidate event classifiers; generates an indicator indicating the accuracy of the sorting strategy; determines that the indicator meets or exceeds a minimum accuracy threshold for the sample; generates a control signal to adjust the operating state of an analysis device included in the particle analyzer, wherein the operating state reflects the sorting strategy; and transmits the control signal to the analysis device to achieve the operating state.

[0022] The analyzing device may include sorting electronics communicatively coupled to the deflection plates. The particle analyzer may be configured to identify a target container suitable for particles based at least in part on a measurement corresponding to the sorting criteria for the target container. Adjusting the operating state may include applying an electrical charge through the deflection plates to direct the particles into the target container.

[0023] The analysis device may include a fluidics system, and adjusting the operating state may include adjusting a pressure applied during an experiment to analyze the sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A functional block diagram of an example of a transformation control system for analyzing and displaying biological events is shown.

[0025] Figure 2A is a schematic diagram of a particle sorter system according to one embodiment described herein.

[0026] Figure 2B is a schematic diagram of a particle sorter system 200 according to one embodiment described herein.

[0027] Figure 3 A functional block diagram of a particle analysis system for computation-based sample analysis and particle characterization is shown.

[0028] Figure 4 is a simplified diagram illustrating an example system for dynamically transforming event data.

[0029] Figure 5 is a simplified diagram showing an example system that uses parametric machine learning transformations to dynamically transform event data.

[0030] Figure 6is a process flow diagram depicting an example of a parametric machine learning transformation method applicable to multidimensional event data. Implementation

[0031] The present invention describes features of a particle sorter that automatically develops a sorting strategy based on a user-selected target event population. The user can select one or more target populations based on measurements including traditional flow measurements and non-traditional parameters such as images and image features. The system then converts the target populations into sorting strategies and configures the sorting hardware accordingly.

[0032] Using user selection to guide the selection of an optimized classifier for sorting solves several problems of user-defined classifiers. One problem involves the difficulty of drawing a gate in a high-dimensional data space. The computing system can simultaneously render two-dimensional or three-dimensional data such as through a graph. The user can select the region in the graph to identify the target population. However, as the dimension increases to more than three, even if it is possible, it is difficult to use the prior art to visualize the gate applicable to the target population and accurately draw the corresponding gate. In addition, even if a given graph can be used to select a population, the gate that truly reflects the target population may also be extremely complex. For example, an asymmetric multi-vertex shape can be used in a high-dimensional space to represent a gate.

[0033] Another issue concerns hindsight bias when selecting new populations of interest. The event sets may be associated with complex biological systems. The user's ability to select may be influenced by previous experience and training. However, the training may not include specific knowledge about parameters that could improve the sorting strategy. In addition, the user may find it difficult to develop a gating strategy based on a large number of parameters.

[0034] Another issue concerns the ability to distinguish events using existing techniques. Because gating relies on a graphical representation of events, in some cases users may not be able to quantify differences between events. It can be difficult to quantify cell “appearance” based on two-dimensional or three-dimensional plots. In addition, populations of interest may inadvertently include subpopulations that are undesirable but hidden among the events of actual interest.

[0035] Another problem relates to the fidelity that can be used to select the event of interest. In some cases, the sortable parameters that quantify the difference to distinguish the events may not be presented to the user in the form of event data. In some cases, composite or transformed parameters can be better used to characterize the event of interest. As another example, parameters describing the appearance of cells may be expensive to calculate or difficult to use for sorting.

[0036] To address these and other issues associated with generating sorting configurations, the present invention describes a two-stage classification process. First, the user can define a target population using a combination of one or more existing methods, such as gating on traditional flow cytometry parameters (e.g., pulse area, width, and height); gating on computationally difficult parameters (e.g., parameters that are difficult or impossible to calculate on commercially available hardware within the waiting time required for droplet sorting, such as image-based parameters (e.g., spot counts, object perimeter, etc.); gating on non-parametric transformations of the data (e.g., gating on the results of dimensionality reduction algorithms such as t-SNE or VerityCEN-SE™); providing example events (e.g., sorted events that are similar to the provided example events); or providing example images (e.g., sorted events that appear similar to the example images).

[0037] In the second stage, the system optimizes the sorting for user-defined target groups by selecting and applying optimized sorting strategies. The strategies may include automatically detecting groups and scoring each event for relevant target groups and non-target groups (computational sorting). The strategies may include using neural networks or other machine learning techniques to approximate transformations (both non-parametric and parametric). The strategies may include projecting target groups from non-sortable parameter space to sortable parameter space. As part of projecting the target group, the system may identify computable parameters that combine to approximate non-computable parameters. The strategies may include automatically extracting features from images or high-dimensional measurements (including, for example, time series waveform data). An example of such automatic feature extraction is an autoencoder neural network for learning relevant image features to be used in sorting decisions.

[0038] Candidate classifiers can be evaluated based on the expected purity or expected sorting yield for the classification. For example, the classification can provide a statistical confidence in the accuracy of the population sorting. The confidence can be used to estimate the purity of the sorted sample (e.g., the proportion of the target population in the sorted material). The performance of the classification algorithm can be characterized using two metrics: precision and recall. Precision is the proportion of positive results that are true positives. Recall is the proportion of true positives that are detected as positive. Maintaining multiple performance metrics may be inefficient. A comprehensive measure that combines precision and recall (e.g., an f-measure) can be used to improve the efficiency and resource utilization of the classification device. For example, a set of example data can be used to calculate the f-measure for each candidate to compare candidate classification strategies. These f-measures can be used to identify the highest performing candidates.

[0039] Once the optimized sorting configuration is generated, the sorting electronics can receive the configuration and adjust operations to sort the samples accordingly. Since the functions of the sorting electronics may vary, the system can generate the sorting strategy based in part on the target sorting electronics. This ensures that the configuration generated in the best way can be applied to the target sorting electronics. For example, the strategy may include transforming event data. Some sorting electronics may not be able to perform certain mathematical operations required for the transformation. In such cases, when requesting a sorting configuration for a limited number of sorting electronics, the candidate classifier set may not include the transformation.

[0040] The terms used herein and specifically set forth below have the following definitions. Unless otherwise defined in this section, all terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0041] As used herein, "system," "apparatus," "device," and "equipment" generally encompass hardware (e.g., mechanical and electronic hardware) and, in some embodiments, related software (e.g., a dedicated computer program for graphics control) components.

[0042] As used herein, "event" or "event data" generally refers to a data packet measured from a single particle (e.g., a cell or synthetic particle). Typically, the data measured from a single particle includes a number of physical measurements, including one or more physical measurements from a detector for measuring light scattering, and at least one parameter or feature derived from fluorescence detected from the particle, such as the fluorescence intensity. Thus, each event is represented as a measurement and a feature vector, where each measured parameter or feature corresponds to a dimension of the data space. In some embodiments, the data measured from a single particle includes images, electrical data, time data, or acoustic data. An event can be associated with an experiment, assay, or sample source (which can be identified by the measurement data).

[0043] As used herein, a "population" or "subpopulation" of particles (e.g., cells or other particles) generally refers to a group of particles having properties (e.g., optical, impedance, or time properties) relative to one or more measured parameters that cause the measured parameter data to form clusters in the data space. Thus, populations are identified as clusters or density regions in the data. In contrast, each data cluster is typically interpreted as corresponding to a particular type of cell or particle population, although clusters corresponding to noise or background are also typically observed. Clusters can be defined in a subset of the dimensions (e.g., a subset relative to the measured parameters), corresponding to populations that differ only in a subset of the measured parameters or features (extracted from the cell or particle measurements).

[0044] As used herein, "gate" generally refers to the boundaries of a classifier that identifies a subset of data of interest. In cytometry, a gate may be associated with a specific set of events of interest. As used herein, "gating" generally refers to the process of classifying a given data set using a defined gate, where the gate can be one or more regions of interest combined with Boolean logic.

[0045] As used herein, an "event" generally refers to an assembled package of data measured from a single particle (e.g., a cell or synthetic particle). Typically, the data measured from a single particle includes many parameters or features, including one or more light scattering parameters or features and at least one other parameter or feature derived from the measured fluorescence. Thus, each event is represented as a vector of parameter and feature measurements, where each measured parameter or feature corresponds to a dimension of the data space.

[0046] As used herein, a "neural network" or "neural network model" can be conceptualized as a network of nodes. The nodes can be organized into layers, where the first layer is an input layer into which data flows. The neural network can also include an output layer, from which transformed data flows out. Each individual node can have multiple inputs and a single output (e.g., an input layer node has only a single input). The output of a node represents a linear combination of the inputs. In other words, the inputs can be multiplied by a relevant constant. The product can be accumulated along a path of nodes with a constant bias. The constant bias or "bias" can represent another degree of freedom that can be adjusted during the training process. For example, for a re-lu based neural network model, the constant bias can be a threshold because it has the ability to reduce the node value below zero, causing the activation function to output zero.

[0047] The resulting value is evaluated using an activation function and used as the output of the node. A node in a given layer of the neural network is connected to each node in an adjacent layer. A neural network can be trained by comparing the desired output of the network with the actual output of the network using a gradient descent algorithm and an error function to minimize the error of the network. The weights of one or more nodes can be adjusted to model the desired results produced by the network.

[0048] Specific examples of various embodiments and systems in which the embodiments and systems may be implemented are described further below.

[0049] Figure 1 A functional block diagram of an example of an adaptive sorting control system for analyzing and displaying biological events is shown. Analysis controller 190 may be configured to implement various processes for controlling the graphical display of biological events.

[0050] The particle analyzer 102 may be configured to collect biological event data. For example, a flow cytometer may generate flow cytometry event data. The particle analyzer 102 may be configured to provide the biological event data to the analysis controller 190. A data communication channel may be included between the particle analyzer 102 and the analysis controller 190. The biological event data may be provided to the analysis controller 190 via the data communication channel.

[0051] Analysis controller 190 may be configured to receive biological event data from particle analyzer 102. The biological event data received from particle analyzer 102 may include flow cytometry event data. Analysis controller 190 may be configured to provide a graphical display including a first graph of biological event data to display device 106. Analysis controller 190 may be further configured to render a region of interest as a gate around the plurality of biological event data shown by display device 106, which is overlaid on the first graph. In some embodiments, the gate may be a logical combination of one or more graphical regions of interest drawn on a single parameter histogram or a bivariate graph.

[0052] Analysis controller 190 may further be configured to display the bio-event data on display device 106 inside the door in a manner different from other events in the bio-event data outside the door. For example, analysis controller 190 may be configured to make the color of the bio-event data contained inside the door different from the color of the bio-event data outside the door. Display device 106 may be implemented in the form of a display, tablet computer, smart phone, or other electronic device configured to display a graphical interface.

[0053] The analysis controller 190 can be configured to receive a door selection signal identifying the door from a first input device. For example, the first input device can be implemented in the form of a mouse 110. The mouse 110 can send a door selection signal to the analysis controller 190 to determine the door to be displayed on the display device 106 or manipulated via the display device 106 (for example, clicking on or in the desired door when the cursor is located there). In some embodiments, the first device can be implemented in the form of a keyboard 108 or other device for providing input signals to the analysis controller 190 (such as a touch screen, a stylus, an optical detector, or a voice recognition system). Some input devices may include multiple input functions. In such embodiments, the input functions can be individually considered as input devices. For example, Figure 1 As shown, the mouse 110 may include a right mouse button and a left mouse button, both of which may generate a trigger event.

[0054] The trigger event may cause the analysis controller 190 to change the way the data is displayed (actually display a portion of the data on the display device 106), or provide input for further processing, such as selecting a target population for particle sorting.

[0055] In some embodiments, analysis controller 190 may be configured to detect when mouse 110 initiates gate selection. Analysis controller 190 may further be configured to automatically modify the graph visualization to facilitate the gating process. The modification may be based on a particular distribution of bio-event data received by analysis controller 190.

[0056] Analysis controller 190 may be coupled to storage device 104. Storage device 104 may be configured to receive and store biological event data from analysis controller 190. Storage device 104 may also be configured to receive and store flow cytometry event data from analysis controller 190. Storage device 104 may also be configured to allow analysis controller 190 to retrieve biological event data, e.g., flow cytometry event data.

[0057] The display device 106 can be configured to receive display data from the analysis controller 190. The display data can include graphs of biological event data and gates outlining portions of the graphs. The display device 106 can be further configured to change the information displayed based on input received from the analysis controller 190 and input received from the particle analyzer 102, the storage device 104, the keyboard 108, and / or the mouse 110.

[0058] In some embodiments, the analysis controller 190 can generate a user interface to receive sample events for sorting. For example, the user interface can include controls for receiving sample events or sample images. Sample events or images or sample gates can be provided before event data for the sample is collected or based on an initial event set for a portion of the sample.

[0059] A common flow sorting technique (referred to as "electrostatic cell sorting") employs droplet sorting, in which a stream or moving column of liquid containing linearly separated particles is broken into droplets, and the droplets containing the particles of interest are charged and deflected into a collection tube by an electric field. Current droplet sorting systems are capable of forming droplets at a rate of 100,000 drops per second in a fluid medium passing through a nozzle with a diameter of less than 100 microns. Droplet sorting typically requires that the droplets detach from the stream at a certain distance from the nozzle tip. The distance is typically on the order of a few millimeters from the nozzle tip, and for an undisturbed fluid medium, the distance is stable and can be maintained by oscillating the nozzle tip at a predetermined frequency and an amplitude that keeps the detachment constant. For example, in some embodiments, the amplitude of a sinusoidal waveform voltage pulse is adjusted at a given frequency to keep the detachment stable and constant.

[0060] Typically, linearly separated particles in the stream are characterized by their passage through an observation point located within a flow cell or cuvette or directly below a nozzle tip. Once a particle is determined to meet one or more of the desired criteria, it is possible to predict when it will reach the droplet breakup point and detach from the stream as a droplet. Ideally, a brief charge is applied to the fluid medium before the droplets containing the selected particles detach from the stream, and then the droplets are grounded immediately after breakup. The droplets to be sorted will remain charged when they detach from the fluid medium, while all other droplets will be uncharged. The charged droplets are deflected laterally from the downward trajectory of the other droplets by the electric field and are collected in a sample tube. The uncharged droplets fall directly into a discharge tube.

[0061] Figure 2A is a schematic diagram of a particle sorter system according to one embodiment described herein. Figure 2A The particle sorter system 250 shown includes deflection plates 252 and 254. Charge is applied via a stream of charge wires in the barbs 256. This creates a particle stream 260 for analysis. The particles can be illuminated with one or more light sources (e.g., lasers) to produce light scattering and generate fluorescence information. The particle information is detected by, for example, sorting electronics or other detection systems ( Figure 2A The deflection plates 252 and 254 can be independently controlled to attract or repel the charged droplets to direct the droplets to a destination collection container (e.g., one of 272, 274, 276, or 278). Figure 2A As shown, the deflector plate can be controlled to direct the particles along the first path 262 toward the container 274 or along the second path 268 toward the container 278. If the particles are not the target particles (e.g., within the specified sorting range, do not show scattering or illumination information), the deflector plate can cause the particles to continue to flow along the flow path 264. Such uncharged droplets can enter a waste container via, for example, an aspirator 270.

[0062] The sorting electronics may be included to begin collecting measurements, receive a fluorescent signal from a particle, and determine how to adjust the deflection plates to sort the particle. Figure 2A An exemplary implementation of the embodiment shown in includes a BD FACSAria™ series flow cytometer commercially available from Becton, Dickinson and Company located in San Jose, California.

[0063] Figure 2B is a schematic diagram of a particle sorter system 200 according to an embodiment described herein. In some embodiments, the particle sorter system 200 is a cell sorter system. Figure 2BAs shown, a droplet formation sensor 202 (e.g., a piezoelectric oscillator) is coupled to a fluid conduit 201 (e.g., a nozzle). Within the fluid conduit 201, a sheath fluid 204 hydrodynamically focuses a sample fluid 206 into a moving liquid column 208 (e.g., a stream). Within the moving liquid column 208, particles (e.g., cells) are aligned to pass through a monitored area 210 (e.g., a laser stream intersection) irradiated by an irradiation source 212 (e.g., a laser). The droplet formation sensor 202 vibrates to break the moving liquid column 208 into a plurality of droplets 209.

[0064] In operation, a detection station 214 (e.g., an event detector) determines when a particle of interest (or cell of interest) passes through the monitored area 210. The detection station 214 feeds a timing circuit 228, which in turn feeds a transient charging circuit 230. At the droplet breakup point, after notification of a timed droplet delay (Δt), a transient charge is applied to the moving liquid column 208 to charge the droplet of interest. The droplet of interest may include one / or more particles or cells to be sorted. The charged droplets may then be sorted by activating a deflection plate (not shown) to deflect the droplets into a container such as a collection tube or a multi-well sample plate, where a well may be associated with a particular droplet of interest. Figure 2B As shown, the droplets are collected in a discharge container 238.

[0065] A detection system 216 (e.g., a droplet boundary detector) is used to automatically determine the phase of the droplet drive signal when a particle of interest passes through the monitored area 210. An exemplary droplet boundary detector is described in U.S. Pat. No. 7,679,039, which is incorporated herein by reference in its entirety. The detection system 216 allows the instrument to accurately calculate the position of each detected particle in the droplet. The detection system 216 can be fed with an amplitude signal 220 and / or a phase 218 signal, which in turn is fed (through an amplifier 222) into an amplitude control circuit 226 and / or a frequency control circuit 224. The amplitude control circuit 226 and / or the frequency control circuit 224 in turn control the droplet formation sensor 202. The amplitude control circuit 226 and / or the frequency control circuit 224 can be included in the control system.

[0066] In some embodiments, the sorting electronics (e.g., detection system 216, detection station 214, processor 240) may be coupled to a memory configured to store detected events and sorting decisions based thereon. The sorting decision may be included in the event data of the particle. In some embodiments, the detection system 216 and the detection station 214 may be implemented in the form of a single detection unit or coupled in a communication manner so that event measurements may be collected by one of the detection system 216 or the detection station 214 and provided to the non-collecting element.

[0067] In some embodiments, one or more of the components described herein applicable to particle sorter system 200 may be used for particle analysis and characterization, whether or not the particles are physically sorted into collection containers. Figure 3 ) can also be used for particle analysis and characterization, whether or not the particles are physically sorted into collection containers. For example, one or more of the components in the particle sorter system 200 or the particle analysis system 300 can be used to group particles or display them in a tree that includes at least three of the groupings described herein.

[0068] Figure 3 A functional block diagram of a particle analysis system for computation-based sample analysis and particle characterization is presented. In some embodiments, particle analysis system 300 is a flow system. Figure 3 The particle analysis system 300 shown may be configured to perform, in whole or in part, the methods described herein, for example, Figure 6 The particle analysis system 300 includes a fluidics system 302. The fluidics system 302 may include or be coupled to a sample tube 310 and a moving liquid column within the sample tube, within which particles 330 (eg, cells) in a sample move along a common sample path 320.

[0069] The particle analysis system 300 includes a detection system 304 configured to collect signals from each particle as each particle passes through one or more detection stations along the common sample path. The detection stations 308 are generally referred to as monitored regions 340 of the common sample path. In some embodiments, detection can include detecting light or one or more other characteristics of the particles 330 as they pass through the monitored regions 340. Figure 3 , a detection station 308 is shown having a monitored area 340. Some embodiments of particle analysis system 300 may include multiple detection stations. In addition, some detection stations may monitor more than one area.

[0070] Each signal is assigned a signal value to form a data point for each particle. As described above, the data can be referred to as event data. The data point can be a multi-dimensional data point, which includes a value for each measured characteristic of the particle. The detection system 304 is configured to collect a series of the data points within a first time interval.

[0071] The particle analysis system 300 also includes a control system 306. The control system 306 may include one or more processors, an amplitude control circuit 226, and / or a frequency control circuit 224 (e.g., Figure 2B306 is operatively associated with the fluidic system 302. The control system 306 is configured to generate a calculated signal frequency for at least a portion of the first time interval based on a Poisson distribution and a number of data points collected by the detection system 804 during the first time interval. The control system 306 is further configured to generate an experimental signal frequency based on the number of data points within the portion of the first time interval. In addition, the control system 306 may compare the experimental signal frequency to the calculated signal frequency or the predetermined signal frequency.

[0072] Figure 4 4 is a diagram illustrating an example event selection system for dynamically identifying event data (e.g., for generating a sorting configuration). Event selection system 400 includes a selection device 420. Selection device 420 includes an event data receiver 422. Event data receiver 422 can receive data from a particle analyzer (e.g., Figure 1 102). In some embodiments, event data 402 may be generated by a particle analyzer by receiving it from, for example, an analysis workstation. For example, a user may provide event data receiver 422 with event data 402 obtained from a particle analyzer. Event data receiver 422 may include a transceiver for wireless communication or a port for connecting to a wired network such as an Ethernet LAN or a port for connecting to a device such as via a universal serial bus or a THUNDERBOLT® connector.

[0073] The event data receiver 422 can provide at least a portion of the event data 402 to an event data processor 424 included in the selection device 420. The event data processor 424 can identify a transformation applicable to the event data. The identification can include detecting a value in the event data, such as an identifier for the assay or experiment. The available transformations can be stored in a data memory 440 accessible to the event data processor 424. The transformation of the event data performed by the event data processor 424 can be a parametric or non-parametric transformation. In some embodiments, the transformation can be specified by a device that provides the event data 402. For example, an analysis workstation can submit a message requesting processing of the event data 402. The message can include a desired transformation (e.g., tSNE). In some embodiments, the event data processing can be guided based on user input. For example, a user can identify a transformation to be applied to the received event data.

[0074] The selection device 420 may include a gate selection 426 unit. The gate selection 426 unit may receive a selection of a destination event from an input device. The selection may be referred to as a gate. The selection may define one or more parameter value ranges applicable to the destination event. The gate selection 426 unit may use the one or more ranges to generate a classifier or other sorting configuration 490. The sorting configuration 490 may be represented as a truth table or decision tree to identify events associated with the gate. As previously described, manually acquiring a gate may be accompanied by errors, and in some cases, manually acquiring a gate may overlook important parameters or potential hardware bottlenecks or efficiencies. Therefore, the initial selection may be used as an example of a desired sorting, which may be adjusted and adapted by further processing (without human intervention) to optimize the retrieval strategy.

[0075] Figure 5 is a diagram illustrating an example system for adaptively generating sorting configurations. Figure 5 The adaptive sorting device 520 in includes receiving selections from a selection device (eg, the selection device 420) and then generating features for an optimized retrieval strategy for evaluating events.

[0076] System 500 includes an adaptive sorting device 520. Adaptive sorting device 520 includes an event data receiver 522. Event data receiver 522 can receive data from a particle analyzer (e.g., Figure 1 In some embodiments, the event data 502 may be generated by the particle analyzer by receiving data from, for example, an analysis workstation. For example, a user may provide the event data 502 obtained from the particle analyzer to the event data receiver 522. The event data receiver 522 may include a transceiver for wireless communication or a network port for connecting to a wired network such as an Ethernet local area network.

[0077] The event data receiver 522 can provide at least a portion of the event data 502 to the selection device 420. The selection device 420 can obtain an example sorting configuration from a user. The example can be provided to the sorting strategy optimizer 524 together with the received event data 502. The sorting strategy optimizer 524 can iteratively generate a sorting strategy that approximates the example sorting configuration. The sorting strategy can include a dynamic pipeline of event data transformation or selection. The available transformations or sorting steps can be stored in a data storage 514 accessible to the sorting strategy optimizer 524. Candidate strategies can include automatically detecting groups and scoring each event for relevant target groups and non-target groups (computational sorting). The strategy can include using neural networks or other machine learning techniques to approximate transformations (non-parametric and parametric). The strategy can include projecting a target group from a non-sortable parameter space to a sortable parameter space. As part of projecting the target group, the system can identify computable parameters that combine to approximate parameters that are not easily computable. The strategy can include automatically extracting features from images or high-dimensional measurements (including, for example, time series waveform data). One example of such automatic feature extraction is an autoencoder neural network for learning relevant image features to be used in sorting decisions.

[0078] In some embodiments, neural networks can be used for feature generation and gating strategy generation. For feature generation, a neural network receives raw image data as input and outputs an indicator of the "appearance" of the image. Other neural networks such as other neural networks receive calculated parameters (e.g., pulse area or height) as input and output new parameters. These new parameters provide a projection from the original parameter space to the new parameter space. For gating strategy generation, a neural network that receives various parameters as input and outputs a single value (which can be used for sorting decisions) can be trained or used to generate sorting decisions.

[0079] Figure 6 is a process flow diagram depicting an example of an adaptive generation method for optimizing a sorting strategy. The method 600 may be performed by an adaptive sorting device (e.g. Figure 5 The adaptive sorting device 520 shown is implemented in whole or in part.

[0080] Method 600 begins at box 602. At box 610, initial event data for a portion of the sample is received. The event data can be collected after activating a particle analyzer to process the portion of the sample. Processing the sample can include measuring properties of the particles, such as graphical, electrical, temporal, or acoustic properties. In some embodiments, the collection of the initial event data can be omitted, and method 600 can proceed from box 602 to box 620. This may be desirable in cases where the sample size is small. In such cases, in order to maintain the number of samples available for sorting, the sorting configuration can be evaluated without consuming any of the sample.

[0081] At box 620, a sorting selection applicable to the sample is received from a communication device. The sorting selection may represent an example of a population of events to be sorted. The sorting selection may include a gate, an example image, or an example event. The sorting selection may be identified relative to the event data of the initial portion of the sample received at box 610. For example, the researcher may draw a polygon on an event data measurement value chart to define a range of data values ​​to be sorted. The polygon may define a gate that may be a sorting criterion or associated with a sorting criterion.

[0082] The sorting options can be used as a reference for developing an optimized sorting configuration. In some embodiments, the sorting options can include information identifying a target sorting instrument or sorting electronics. As previously described, different hardware can have different functions for implementing a sorting configuration. In order to ensure that the sorting configuration is suitable for the target hardware, the identity of the instrument can be considered.

[0083] At box 630, candidate event classifiers suitable for the sample can be identified. The identification can include selecting an event classifier from a data storage device. The event classifier can include a neural network model, an event data transformation, an autoencoder, or other machine-implemented element for evaluating event data. The selection can be based in part on the sample type (e.g., blood, urine, tissue, etc.). The selection can be based in part on the particle analyzer that will be used to process the sample. The selection can be based in part on the sorting selection received from the communication device. For example, if the sorting selection includes an image, a graph-based classifier can be selected as a candidate event classifier. As another example, if the event distribution in the sorting selection is characterized by a statistically regular distribution, it may be necessary to use a classifier based on Mahalanobis distance.

[0084] In some embodiments, a user interface may be provided to collect information specifying classifiers included in the candidate event classifiers. In some embodiments, the system may consider a set of parameters and classifiers defined in a data store. The classifiers may be filtered based on, for example, the manner in which the sample event is identified by a user. When identifying a classifier, the historical record identification information may be used to identify a general workflow for bundling the classifier with parameters that are selected together for the experiment, the target particle, the particle analyzer that generated the data, the experiment, or other detectable characteristics of the data generated during the experiment.

[0085] At box 640, one or more candidate event classifiers may be used to generate a sorting strategy. Generating a sorting strategy may include ranking one or more candidate event classifiers to form a pipeline for processing event data. Generating the sorting strategy may include using the user's example gating strategy to inform the appropriate connection of different classifiers and / or parameters. For example, if the user uses a hierarchical gate to derive their example population, we can use a similar hierarchy. Another example is that if the user uses a transformed space at any point in determining their example data, the system can detect the transformation and generate an approximation of the transformation for the sorting strategy. In addition, another example is to detect a large number of hierarchical gates in an attempt to draw manifolds in a high-dimensional space and use tools such as relationship-preserving transformations or statistical models such as Mahalanobis distances. When identifying a sorting strategy, historical record strategy information can be used to identify a general workflow that ranks specific identifiers for experiments, target particles, particle analyzers that generate the data, experiments, or other detectable characteristics of data generated during experiments.

[0086] The sorting strategy generated at box 640 can be evaluated using an indicator. At box 650, an indicator indicating the accuracy of the sorting strategy is generated. The indicator can represent the accuracy (e.g., purity) of the sorted events from the sample. The indicator can be generated based on the confidence of the classifier included in the sorting strategy. In some embodiments, the indicator can be generated based on the comparison of the sorting selection with the sorting configuration generated by the sorting strategy. One way to generate the indicator is to use an F-measure for the sorting strategy, wherein the precision indicates the level of sorting purity, and the recall indicates the number or yield of the sorted samples. For example, the example events provided by the user can be divided into "training" and "testing" subsets. The division can be performed based on a pseudo-random selection of a portion of the events. The training subset can be used to train a variety of candidate gating strategies. These gating strategies can then be used to evaluate the test subset. The results of this test will be used to generate an F-measure.

[0087] At block 660, it is determined whether the indicator applicable to the sorting strategy corresponds to a threshold. The threshold may be a predetermined configuration value indicating a minimum purity or yield of the sorting strategy. If the result of the determination at block 660 is positive, then the generated sorting strategy may be considered appropriate for the sample. In such a case, method 600 proceeds to block 670.

[0088] At block 670, sorting electronics (e.g., sorting circuitry) may be configured using the sorting strategy generated at block 640. Configuration of the sorting electronics may include storing the models or transformations included in the strategy in a storage location accessible to the sorting electronics. Event data may then be processed using the sorting strategy to be evaluated against the sorting criteria included in the sorting strategy.

[0089] At block 680, the analyzer may evaluate and sort the remainder of the sample using the configured sorting electronics. As new event measurements are collected, the measurements may be processed in real time using the configured sorting electronics and sorted into designated containers according to the sorting configuration. For example, a deflection plate of the particle analyzer may be activated to direct the particles of interest into a designated collection tube.

[0090] Method 600 ends at box 690. However, it should be understood that method 600 can be repeated for other events, samples, or experiments. In some embodiments, it is desirable to generate a new sorting strategy to accommodate any variation within the sample or to account for changes in the source of the sample. For example, in a therapeutic setting, a biological sample may be collected during the administration of a drug or other compound. The sorting strategy may need to be adjusted to account for the presence of the drug or compound after administration or unexpected changes compared to the initial example selection used to guide the generation of the adaptive strategy. In such cases, given that the original retrieval strategy has been trained, the sorting strategy may be regenerated in part based on the collected data. For example, the initial strategy may have identified a normal distribution of events, but the actual event data collected for the sample may indicate an event with a non-normal distribution.

[0091] The term "determining" as used herein encompasses a variety of activities. For example, "determining" may include calculating, computer computing, processing, deriving, investigating, looking up (e.g., looking up in a table, database, or another data structure), ascertaining, etc. In addition, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), etc. In addition, "determining" may include resolving, selecting, choosing, establishing, etc.

[0092] As used herein, the term "providing" encompasses a variety of activities. For example, "providing" may include storing a value at a location on a storage device that facilitates subsequent retrieval, sending a value directly to a recipient via at least one wired or wireless communication medium, sending or storing a reference to a value, etc. "Providing" may also include encoding, decoding, encryption, decryption, validation, verification, etc., performed via hardware elements.

[0093] The terms "selectively" or "selective" as used herein may encompass a variety of activities. For example, a "selective" process may include determining an option from a plurality of options. A "selective" process may include one or more of the following: dynamically determined inputs, preconfigured inputs, or user-initiated inputs for determination. In some embodiments, n-input switching may be included to provide selective functionality, where n is the number of inputs used to make a selection.

[0094] The term "message" as used herein encompasses various formats for conveying (e.g., sending or receiving) information. A message may include a machine-readable aggregation of information, such as an XML document, a fixed field message, a comma-delimited message, etc. In some embodiments, a message may include a signal for transmitting one or more information representations. Although described in singular form, it should be understood that a message may be composed of multiple parts, sent, stored, received, etc.

[0095] As used herein, "user interface" (also referred to as interactive user interface, graphical user interface or UI) may refer to a web-based interface including data fields, buttons or other interactive controls for receiving input signals or providing electronic information or providing information to a user in response to any received input signals. The UI may be implemented in whole or in part using technologies such as Hypertext Markup Language (HTML), JAVASCRIPT™, FLASH™, JAVA™, .NET™, WINDOWS OS™, macOS™, Web services and Rich Site Summary (RSS). In some embodiments, the UI may be included in an independent client (e.g., a fat client, a fat client) configured to communicate (e.g., send or receive data) according to one or more of the aspects described.

[0096] As used herein, a "data storage area" may be embodied in a hard disk drive, solid-state memory, and / or any other type of non-transitory computer-readable storage medium that can access a device or be accessed by the device, including an access device, a server, or other computing device, etc. As is known in the art, the data storage area may also or alternatively be distributed or partitioned across multiple local and / or remote storage devices without departing from the scope of the present invention. In other embodiments, the data storage area may include or be embodied in a data storage network service.

[0097] As used herein, a phrase referring to "at least one" of a list of items refers to any combination of the items, including individual members. For example, "at least one of the following items: a, b, or c" is intended to cover: a, b, c, ab, ac, bc, and abc.

[0098] Those skilled in the art will appreciate that any of a variety of different technologies and techniques may be used to represent information, messages, and signals. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be mentioned throughout the above specification may be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or optical particles, or any combination thereof.

[0099] It will be further understood by those skilled in the art that the various illustrative logic blocks, modules, circuits and algorithmic steps described in conjunction with the embodiments disclosed herein can be used as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits and steps have been generally described above with respect to their functions. Whether this function is implemented in the form of hardware or software depends on specific applications and design constraints for the entire system. The technician can perform the functions in various ways for each specific application, but this execution decision should not be interpreted as causing departure from the scope of the present invention.

[0100] The technology described herein can be implemented in the form of hardware, software, firmware, or any combination thereof. The technology can be implemented in any of a variety of devices, such as a specially programmed event processing computer, a wireless communication device, or an integrated circuit device. Any function described as a module or component can be performed together in an integrated logic device, or separately in a discrete but interoperable logic device. If implemented in software form, the technology can be implemented at least in part by a computer-readable data storage medium, which includes instructions, and when the instructions are executed, one or more of the above methods are executed. The computer-readable data storage medium may constitute a part of a computer program product, which may include packaging materials. The computer-readable medium may include a memory or a data storage medium, such as a random access memory (RAM) (e.g., a synchronous dynamic random access memory (SDRAM)), a read-only memory (ROM), a non-volatile random access memory (NVRAM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic or optical data storage medium, etc. The computer-readable medium may be a non-temporary storage medium. Alternatively or additionally, the techniques may be implemented at least in part by a computer-readable communication medium that carries or communicates program code in the form of instructions or data structures and that can be accessed, read, and / or executed by a computing device, such as a propagated signal or wave.

[0101] The program code can be executed by a specially programmed adaptive sorting strategy processor, which can include one or more processors, for example, one or more digital signal processors (DSPs), configurable microprocessors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other equivalent integrated or discrete logic circuit systems. The graphics processor can be specially configured to perform any of the techniques described in the present invention. A combination of computing devices (e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in combination with a DSP core, or any other such configuration in at least part of the data connection) can perform one or more of the functions described. In some aspects, the functions described herein can be provided in a dedicated software module or hardware module configured for encoding and decoding, or incorporated into a dedicated sorting control card.

[0102] The method disclosed herein includes one or more steps or actions for implementing the method. The method steps and / or actions may be interchangeable with each other without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims.

[0103] Various embodiments of the invention are described herein. These and other embodiments are within the scope of the following illustrative claims.

Claims

1. A computer-implemented method comprising: controlled by one or more processing devices, receiving, with the particle analyzer, from a communication device, a reference sorting standard associated with a population of interest in the sample; generating a sorting strategy that is similar to the reference sorting standard; as well as The particle analyzer is configured to classify the particles of the sample based at least in part on the multi-dimensional measurements of the particles and a sorting strategy. 2 . The computer-implemented method of claim 1 , wherein generating a sorting strategy that approximates the reference sorting criterion comprises generating a sorting strategy in an iterative manner.

3. A computer-implemented method according to claim 1 or 2, wherein the sorting strategy comprises a dynamic pipeline of data transformations.

4. The computer-implemented method of any one of the preceding claims, wherein generating a sorting strategy that approximates the reference sorting criterion comprises training a neural network based at least in part on the reference sorting criterion.

5. A computer-implemented method according to any one of the preceding claims, wherein the reference sorting criteria comprises gate information identifying a range of measurement values ​​for classifying the particles.

6. A computer-implemented method according to any one of the preceding claims, wherein the reference sorting standard comprises an image showing reference particles to be collected from the sample.

7. The computer-implemented method of any preceding claim, further comprising: transmitting a sorter configuration to the particle analyzer, wherein the sorter configuration represents the sorting strategy; and Sorting circuitry is adjusted based at least in part on the sorter configuration.

8. The computer-implemented method of any of the preceding claims, wherein the multi-dimensional measurements received from the particle analyzer include measurements of light emitted by the particles in the form of fluorescence.

9. The computer-implemented method of claim 8, wherein the light emitted by the particle in the form of fluorescence comprises light emitted by an antibody bound to the particle in the form of fluorescence.

10. A system comprising: one or more processing devices; and A computer-readable storage medium containing instructions that, when executed by the one or more processing devices, cause the system to: receiving, with the particle analyzer, from a communication device, a reference sorting standard associated with a population of interest in the sample; generating a sorting strategy that is similar to the reference sorting standard; as well as The particle analyzer is configured to classify the particles of the sample based at least in part on the multi-dimensional measurements of the particles and a sorting strategy.

11. The system of claim 10, wherein generating a sorting strategy that approximates the reference sorting criterion comprises generating a sorting strategy in an iterative manner.

12. The system of claim 10 or 11, wherein the sorting strategy comprises a dynamic pipeline of data transformations.

13. The system of any one of claims 10 to 12, wherein generating a sorting strategy that approximates the reference sorting criterion comprises training a neural network based at least in part on the reference sorting criterion.

14. The system of any one of claims 10 to 13, wherein the reference sorting criterion comprises gate information identifying a range of measurement values ​​for classifying the particles.

15. The system of any one of claims 10 to 14, wherein the reference sorting standard comprises an image showing reference particles to be collected from the sample.

16. The system of any one of claims 10 to 15, wherein the computer-readable storage medium contains instructions that, when executed by the one or more processing devices, further cause the system to: transmitting a sorter configuration to the particle analyzer, wherein the sorter configuration represents the sorting strategy; and Sorting circuitry is adjusted based at least in part on the sorter configuration.

17. The system of claim 16, wherein the sorting circuit comprises a field programmable gate array.

18. The system of any one of claims 10 to 17, wherein the multi-dimensional measurements received from the particle analyzer include measurements of light emitted by the particles in the form of fluorescence.

19. The system of claim 18, wherein the light emitted in the form of fluorescence by the particle comprises light emitted in the form of fluorescence by antibodies bound to the particle.

Images