Cell characterisation

Abstract

A method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell, the method comprising: determining cell boundary profile data associated with a cell subject to steady or transient deformation; inputting the cell boundary profile data into a machine learning model; and using the machine learning model to determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of the cell based on the cell boundary profile data.

Description

CELL CHARACTERISATIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims priority from GB patent application 2403011.6 filed 1March 2024, the entire contents of which are hereby incorporated by reference.FIELD OF THE INVENTION

[0002] The invention relates to cell analysis and sorting. In particular, but not exclusively, the invention relates to methods and apparatuses for determining intrinsic mechanical properties and sizes of subcellular components of cells in order to classify and sort the cells.BACKGROUND

[0003] Cells are the fundamental units of life. It is known that the intrinsic mechanical properties of cells, such as elasticity and viscosity, influence the cells’ differentiation, growth, migration, and survival. Accordingly, these properties are promising biomarkers for diagnosis and prognosis of diseases, including cancer, malaria, sepsis, and vascular disorders.

[0004] However, it is difficult to measure the intrinsic mechanical properties of cells due to their small size and fragility. Further, clinical applications require huge quantities of cells (e.g., thousands to millions) to be measured. Accordingly, obtaining measurements of the intrinsic mechanical properties of cells is extremely difficult, time-consuming and labour-intensive.

[0005] For characterisation of the intrinsic mechanical properties of cells, one needs to deform the individual cells with a well-defined force, measure the deformation, and then fit the forcedeformation data to a physical model to inversely infer the values of the mechanical properties. Classical characterisation methods, such as atomic force microscopy (AFM), micropipette aspiration, optical stretcher, and magnetic twisting cytometry often suffer from low throughput rates that are typically limited to 10-1000 cells per hour. As a result, characterising millions of cells remains unpractical.

[0006] Previously, in order to address the low-throughput problem, hydrodynamic approaches have been proposed, where cells were flowed through microfluidic channels, and were deformed by the fluid flow at a throughput rate of up to thousands of cells per second. A high-speed camera was used to record the cell deformation, and the images were subsequently analysed to determine the deformability of the cells. Such methods are described, for example, in US Patent Number US 10,571,387 B2 and US Patent Application Number US 2022 / 0026340 Al.

[0007] Whilst these methods refer to mechanical properties of cell, they are in fact directed to quantifying the cell deformability, represented by quantitative morphological parameters,including the deformation index (in US 10,571,387 B2), elongation ratio and circularity (US 2022 / 0026340 Al). Those quantitative parameters are not intrinsic mechanical properties and their values change with many experimental parameters such as the flow speed, and the viscosity of the cell suspension medium. For the same cell, its deformation will be larger if being flowed through a channel at a higher speed, or in a medium with a higher viscosity. Therefore, using such methods means that one must use the same set of flow parameters to operate the experiments, otherwise cell deformability obtained from different experiments cannot be compared with each other. Only being able to operate experiments under one set of flow parameters is a significant limitation to the number of cells that can be processed. This is because cells are extremely heterogeneous, which means that in order to achieve sufficient deformation, the flow speed or suspension viscosity needs to span a wide range depends on the intrinsic elasticity and viscosity of the cells. Accordingly, known methods do not address high throughput measurements of intrinsic mechanical properties of cells.

[0008] Further, whilst cell deformability has been proposed as a criterion for cell classification (e g., in US 2022 / 0026340 Al), due to the significant heterogeneity of cells even within the same cell type, it is doubtful how well the deformability of the whole cell would be able to distinguish cells of different types. For example, the Young’s modulus (a parameter for intrinsic elasticity) of breast cancer and leukaemia cells spans 0.1-0.8 kPa and 0.02-1.4 kPa, respectively, and both overlap with the 0.07-0.2 kPa of neutrophils (the most numerous type of white blood cells)

[0009] It is an object of the disclosure to at least partly address one or more of the shortcomings in the prior art.SUMMARY OF THE INVENTION

[0010] According to an aspect of the invention, there is provided a method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell, the method comprising: determining cell boundary profile data associated with a cell subject to steady or transient deformation; inputting the cell boundary profile data into a machine learning model; and using the machine learning model to determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of the cell based on the cell boundary profile data.

[0011] There is also provided a computing device configured to perform the method and a computer readable medium comprising instructions that, when executed by a computer, cause the computer to perform the method.

[0012] There is also provided a system for cell analysis, the system comprising: a microfluidic arrangement for controllably subjecting one or more cells to steady and / or transient deformationconditions; an imaging apparatus for imaging the one or more cells subject to steady and / or transient deformation conditions; and a computing device configured to perform the method.

[0013] Advantageously, real-time imaging of the flow-induced deformation of cells is used to provide cell boundary profile data in order to determine intrinsic mechanical properties and / or size of subcellular components of a cell. Beneficially, cells are processed with a high-throughput such that cell boundary profile data reveals detailed information regarding the subcellular properties of cells without the need for measuring the subcellular components directly. For example, one can characterise the size and intrinsic mechanical properties of the cell nucleus without directly visualising or touching it.

[0014] Optionally, the method further comprises: classifying the cell type and / or determining the cell status based on the determined one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell. Advantageously, determination of one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell enables further classification of the cell type and / or determination of the cell status based on a high-throughput measurement of the cell boundary. Beneficially, the efficient processing of cell boundary profile data serves as a label-free biomarker for the efficient processing of cells.

[0015] Optionally, determining cell boundary profile data associated with a cell subject to steady or transient deformation comprises: applying one or more steady and / or transient deformation conditions to cell; and capturing one or more images of the cell associated with the one or more steady and / or transient deformation conditions. Optionally, applying one or more steady and / or transient deformation conditions to the cell comprises: directing the cell to flow through one or more microchannels comprising a geometry configured to induce cell deformation, optionally wherein the one or more microchannels comprises a straight channel geometry, a converging channel geometry, a diverging channel geometry, a wavy channel geometry, a cross-slot channel geometry or a branched channel geometry, optionally wherein at least a portion of the channel has a cross-sectional profile that is square, rectangular, circular, semi-circular, elliptical or semielliptical. Beneficially, the use of flow-induced deformation enables the precise control of flow fields to cells whilst also enabling high-throughput. Advantageously, a wide range of deformation conditions are straightforwardly applied, thereby enabling the response of a wide range of heterogeneous cells to be determined.

[0016] Optionally, determining cell boundary profile data comprises: determining one or more 1-dimensional vectors of coordinates of cell boundary nodes based on the one or more images. Optionally, determining cell boundary profile data comprises: binarizing at least a portion of the one or more images to provide binarized data; extracting cell boundary data from the binarized data; determining nodes associated with the cell boundary data; and determining a coordinatevector based on the determined nodes. Optionally, extracting the cell boundary data comprises: fitting a cell boundary curve to the binarized data to provide cell boundary data corresponding to the fitted cell boundary curve, optionally wherein fitting a cell boundary curve to the binarized data comprises applying a piecewise second-order polynomial function to the binarized data to provide cell boundary data corresponding to a fitted cell boundary curve. Optionally, determining nodes associated with the cell boundary data comprises: sampling data points with an equal arc-length separation function corresponding to the fitted cell boundary curve. Advantageously, images of cells are reduced to vector coordinates, which enables the efficient processing of large amounts of data whilst simultaneously enabling detailed subcellular information to be determined.

[0017] Optionally, the machine learning model has been trained using simulated cell boundary profile data corresponding to the steady and / or transient deformation of a cell subject to one or more steady and / or transient deformation conditions, wherein the simulated cell boundary profile data is based on a mechanistic model of the cell comprising one or more intrinsic mechanical properties of one or more subcellular components of the cell. Advantageously, efficient mechanistic models are used to provide high-fidelity simulations without needing to collect experimental data, which is difficult and time consuming.

[0018] Optionally, the method further comprises: sorting the cell by selectively directing the cell in one of a plurality of directions based on the classified cell type and / or cell status by applying a physical force to directly drive the cell, or using a physical force to drive a fluid flow to direct the cell, optionally wherein the physical force is generated by at least one of a piezoelectric component, an electrostatic approach, an optomechanical approach, by using compressed air, and / or valves. Beneficially, high-throughput real-time imaging and determination of one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell means that cells can also be sorted based on cell types and statuses derived from the underlying measurements.

[0019] Optionally, the cell is obtained from a sample of a subject, and the method further comprises: prognosing or diagnosing a condition of the subject and / or monitoring the efficacy of a drug based on the classified cell type and / or cell status of one or more cells. Optionally, the sample is a blood sample or the sample is a cell suspension, optionally wherein the cell suspension comprises cells obtained from a patient tissue by physically grinding or biochemically dissolving the patient tissue and suspending the physically ground or biochemically dissolved patient tissue in a liquid medium. Optionally, the condition comprises at least one of cancer, cardiovascular disease, diabetes and sepsis, optionally wherein the method comprises using the cell status to monitor the efficacy of drugs for treating the condition.Advantageously, a label-free biomarker is provided by determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell from the cell boundary profiles, for the efficient determination of diseases and treatment efficacy.

[0020] There is also provided a method of training a machine learning model for determining one or more intrinsic mechanical properties and / or sizes of one or more subcellular components of a cell, the method comprising: receiving simulated cell boundary profile data corresponding to the steady and / or transient deformation of a cell subject to one or more steady and / or transient deformation conditions, wherein the simulated cell boundary profile data is based on a mechanistic model of the cell comprising one or more intrinsic mechanical properties of one or more subcellular components for a plurality of different parameters; and training the machine learning model to predict intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on cell boundary profile of the cell. Optionally, training the machine learning model comprises supervised training. Advantageously, an appropriately trained machine learning model provides an ultralow latency tool for the indirect determination of the intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on cell boundary profile of the cell.

[0021] Optionally, the method comprises generating simulated cell boundary profile data by: determining a mechanistic model of a cell based on the intrinsic mechanical properties of one or more subcellular components of the cell; using the mechanistic model to simulate cell boundary profile data associated with one or more steady and / or transient deformation conditions and / or cell properties, optionally wherein the one or more steady state and / or transient deformation conditions comprise fluid medium rheology and flow speed and / or wherein the cell properties comprise cell size, cell shape and / or intrinsic mechanical properties of one or more subcellular components of the cell. Advantageously, efficient mechanistic models are used to provide high- fidelity simulations without needing to collect experimental data, which is technically challenging, time consuming, and expensive.

[0022] Optionally, the subcellular components include at least one of: a cell membrane, a cytoplasm and a nucleus. Optionally, the one or more intrinsic mechanical properties of the subcellular components comprise at least one of: a Young’s modulus, a Poisson ratio and a viscosity. Beneficially, such features are used to determine information regarding a cell from a cell boundary.

[0023] Optionally, the machine learning model comprises at least one neural network, optionally wherein the neural network is a multi-layer perceptron, a support vector machine, a random forest or a long short-term memory network.

[0024] Optionally, the cell type corresponds to a normal human cell type or a diseased cell type, optionally wherein the diseased cell type is a cancerous cell type, optionally wherein the cell type is a prostate cancer cell type or a leukaemia cell type.

[0025] Optionally, the cell status corresponds to a normal / healthy, or a hardened / softened, or an activated / inactivated, or a cell in a specific cell development cycle.

[0026] There is also provided a system for cell sorting, the system comprising: a microfluidic arrangement for controllably subjecting one or more cells to steady and / or transient deformation conditions; an imaging apparatus for imaging the one or more cells subject to steady and / or transient deformation conditions; a computing device configured to perform the method of sorting the cell by selectively directing the cell in one of a plurality of directions based on the classified cell type and / or cell status by applying a physical force to directly drive the cell, or using a physical force to drive a fluid flow to direct the cell; and a cell sorting arrangement for selectively directing the one or more cells in one of a plurality of directions.

[0027] There is also provided a machine learning model for determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell, wherein the machine learning model is configured to: receive cell boundary profile data corresponding to a steadily and / or transiently deformed cell; and determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of the transiently deformed cell based on cell boundary profile data. Optionally, the machine learning model is further configured to: classify the cell type and / or cell status based on the determined one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell. Optionally, the machine learning model comprises at least one neural network, optionally wherein the neural network is a multi-layer perceptron, a support vector machine, a random forest or a long short-term memory network.

[0028] Further aspects of the invention will be apparent from the description and the appended claims.DETAILED DESCRIPTION OF AN EMBODIMENT

[0029] A detailed description of embodiments is described, by way of example only, with reference to the figures, in which:Figure 1 shows a process flow for determining, classifying and / or sorting cells;Figure 2 shows a process flow for training a machine learning model;Figure 3 shows a system for cell analysis;Figures 4A to 4B show different microfluidic channel geometries for providing transient cell deformation;Figure 5A shows a process flow for extracting cell boundary profile data from an image; Figure 5B illustrates steps of the process of Figure 5A;Figure 6A shows subcellular components of a cell for a mechanistic model;Figure 6B shows simulated cell boundary profile data;Figures 7A and 7B show images of cells under different flow-induced deformation conditions with corresponding simulated cell boundary profile data;Figure 8A shows a plot of cell membrane elasticity as a function of cell diameter for prostate cancer PC3 cells;Figure 8B shows a plot of cell membrane elasticity as a function of cell diameter for leukaemia K562 cells;Figure 8C shows plots of a statistical distribution as a function of cell diameter for prostate cancer PC3 cells and leukaemia K562 cells;Figure 8D shows plots of a statistical distribution as a function of cell membrane elasticity for prostate cancer PC3 cells and leukaemia K562 cells;Figures 9A to 9C show plots of distributions of cell membrane elasticity and cell diameter for freshly isolated peripheral blood mononucleate cells for healthy participants, presurgery prostate cancer patients and post-surgery prostate cancer patients, respectively;Figure 10 shows simulated (curves) and experimental (dots) plots 1000 of the time evolutions of the cell deformation parameter when the cell is flowing through a cross-slot microchannel. Simulated results have covered cells with different sizes of nucleus; and Figures 11 A and B show plots of cell membrane viscosity as a function of cell diameter for prostate cancer PC3 cells and leukaemia K562 cells respectively.

[0030] Classification of the type and status of cells and the subsequent sorting of cells of interests from a heterogeneous population has widespread biological and biomedical applications. The current golden standard technology is fluorescence activated cell sorting (FACS), which uses molecular biology and fluorescent tags, that are expensive and often cause cell damage, which is detrimental for downstream applications such as cell therapy. Whilst label-free approaches have been proposed for cell classification in recent years, existing high throughput methods can only measure the apparent deformability of the whole cell. That means, a cell is considered as a homogeneous body that does not have any subcellular structure. None of the high-throughput methods can measure the intrinsic mechanical properties of subcellular components including the membrane, cytoplasm and nucleus.

[0031] A method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on cell boundary profile data is described herein.Where properties of those subcellular components, such as the elasticity and size of cell nucleus, are closely related to diseases such as cancer, advantageously, determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on the cell boundary profile data enables real-time high-throughput classification of cell types and determination of cell statuses to identify disease types and statuses. Beneficially, the methods described herein enable the screening of large-populations of human cells of interest and characterisation of the intrinsic mechanical properties (e.g., Young’s modulus, Poisson ratio, viscosity) of the subcellular cell components, including the membrane, cytoplasm and nucleus, in order to uncover and thoroughly understand the differences of different cell types at the subcellular-level. This provides improved cell analysis in an elegant and efficient way that cannot be achieved using known methods.

[0032] Figure 1 shows a process flow SI 00 of a method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell. The process is initiated at step SI 02, for example by a user of cell analysis system, such as the cell analysis system described with reference to Figure 3. Figure 3 shows a system for cell analysis 300, that is configured to perform the process flow S100. The system for cell analysis 300 comprises a microfluidic arrangement 302 for controllably subjecting one or more cells to steady and / or transient deformation conditions, an imaging apparatus 306 for imaging the one or more cells subject to steady and / or transient deformation conditions, and a computing device 308 configured to perform a method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell. The system for cell analysis 300 uses cell boundary profile data 314 as an input to a machine learning model 316 that is trained on a high-fidelity mechanistic model 318 of a cell in order to output data 320, 322, 324, as described in further detail below. Whilst the process flow S100 is described with reference to the system for cell analysis 300 described with reference to Figure 3, in further examples, additional and / or alternative apparatus is used to perform the process flow S100.

[0033] Beneficially, the system for cell analysis 300 is a high-throughput platform that can accurately and simultaneously characterise multiple intrinsic properties of the membrane, cytoplasm and nucleus of living cells in real time (for example within 0.0001 seconds per cell) with a throughput rate of up to 10000 cells per second, which is five orders of magnitude faster than known technologies. Advantageously, the determined intrinsic mechanical properties can be used as label-free biomarkers in order to classify cell type and / or status with a high accuracy of 90% and higher.

[0034] In order to determine intrinsic mechanical properties and / or size of subcellular components of a cell, cell boundary profile data 314 is used. Where cell boundary profile data314 is provided, for example where experimental processing of a sample of cells is performed remotely and the process S100 is initiated at step S102 in response to cell boundary profile data 314 being received by a user of a computing device 308, the process moves to step SI 10, where the cell boundary profile data 314 is input into a machine learning model 316, as described herein.

[0035] Where cell boundary profile data 314 is not yet provided, optionally, following initiation of the process SI 00 at step SI 02, for example in response to receiving a sample, steps SI 04 to SI 08 are performed in order to obtain cell boundary profile data 314 for use as an input for a machine learning model 316. Cell profile boundary data 314 is determined by applying one or more steady and / or transient deformation conditions to a cell and capturing one or more images of the cell associated with the one or more steady and / or transient deformation conditions. In order to subject the one or more cells to one or more steady and / or transient deformation conditions, the process flow moves to step S104 where flow-induced cell deformation is initiated. Flow-induced cell deformation may be initiated in response to receiving a sample, such as a blood sample. Additionally, or alternatively, the sample is a cell suspension. Optionally, a cell suspension is obtained from patient tissue by physically grinding or biochemically dissolving the patient tissue and suspending the physically ground or biochemically dissolved patient tissue in a liquid medium. Additionally, or alternatively, the sample is any appropriate sample that provides cells in such a way that they can be subjected to flow-induced cell deformation.

[0036] Flow-induced cell deformation is achieved using the microfluidic arrangement 302, which enables high-throughput, controllable, flow-induced cell deformation. The microfluidic arrangement 302 is a platform that can be used to flow and deform from tens to tens of thousands of cells per second with precisely controlled flow conditions. Figure 3 shows an example of a cell 303 that is directed through a microchannel 304 of the microfluidic arrangement 302. In order to direct cells through the microchannel 304, one or more cells are introduced into the microfluidic arrangement 302 via an inlet and directed by the flow of the medium in which they are suspended, as illustrated by the arrows in the microchannel 304 of Figure 3. The cell 303 is shown in different positions 303a, 303b, 303c, 303d, 303e, 303f corresponding to different times during the flow of the cell 303 through the microchannel 304. The geometry of the microchannel 304 creates a flow field with a spatial variation that results in local forces on cells flowing through the microchannel 304. Therefore, appropriate control of the geometry of the microchannel 304 and fluid flow through the microchannel 304 means that forces on a cell 303 flowing through the microchannel 304 can be precisely controlled and known at different positions 303a, 303b, 303c, 303d, 303e, 303f. Transient deformation conditions are thereforeprovided by the different forces applied to a cell as the cell flows through the microchannel 304. Steady deformation conditions can also be applied such that the flow of fluid through the microchannel 304 provides a constant known deformation to a cell at a known position in the microfluidic arrangement 302. Consequently, a great deal of information is provided in the form of a cell’s individual response to the forces applied under known deformation conditions.

[0037] In the example of Figure 3, the geometry of the microchannel 304 is a branched geometry that induces a deformation in the cell 303. In further examples the microchannel 304 is formed with any suitable geometry to induce transient and / or steady-state deformation of a cell in a microfluidic arrangement 302 by introducing a flow field with a spatial variation so that cells can undergo transient deformation when flowing through the microchannel 304. Whilst the flow- induced deformation of cells is described with reference to the microfluidic arrangement 302 of Figure 3, in further examples, any appropriate alternative and / or additional microfluidic apparatus is used in order to enable cells to be subjected to flow-induced deformation under known conditions, such that their deformation can be recorded in one or more images and associated with the known deformation conditions.

[0038] By way of example, Figures 4A to 4D show possible microchannel 304 geometries for subjecting a cell to deformation. Figure 4A shows a converging channel geometry 404A. Figure 4B shows wavy channel geometries in Forward 404B, Symmetric 404B’ and Reverse 404B” arrangements. Figure 4C shows a cross-slot channel geometry 404C. Figure 4D shows a branched channel geometry 404D. In further examples, the microchannel 304 has a sinusoidal channel geometry, a straight channel geometry and / or a diverging channel geometry. The cross- sectional profile of the microchannel 304 has any appropriate form for subjecting one or more cells to deformation conditions. For example, at least a portion of the microchannel 304 may have a cross-sectional profile that is square, rectangular, circular, semi-circular, elliptical or semi-elliptical. In an example, the microchannel 304 has different cross-sectional profiles at different portions of the microchannel 304.

[0039] Once flow-induced cell deformation has been applied at step SI 04, the process SI 00 moves to step SI 06, where high-speed imaging of the cell 303 is performed in order to capture one or more images 312 of a cell associate with one or more steady and / or transient deformation conditions. Whilst the high-speed imaging of a cell is shown in a separate step S106 to the flow- induced cell deformation step SI 04, it will be understood that the steps SI 04, SI 06 can be performed simultaneously as part of a continuous process of imaging cells flowing through the microfluidic arrangement 302. The transient deformation of the cell 303 is captured by an imaging apparatus 306 to provide one or more images 312 of a cell corresponding to knowable deformation conditions. Accordingly, where the cell 303 is flowing through the microchannel304, the positions of the cell 303a, 303b, 303c, 303d, 303e, 303f can be correlated with forces associated with local flow field. Advantageously, the same or nearby positions within the microfluidic channel 304 may be used for a high number of measurements of different cells subject to the same flow-induced deformation conditions and different positions within the microfluidic channel 304 may be used for multiple measurements of the same cell subject to different flow-induced deformation conditions.

[0040] The imaging apparatus 306 is a high-speed imaging apparatus that enables real-time data capture. Images 312 captured by the imaging apparatus 306 are bright-field images. In further examples, alternatively or additionally, the images 312 captured by the imaging apparatus 306 are fluorescence images. In further examples, additionally or alternatively, the images 312 are any appropriate type of image for extracting cell boundary data in accordance with the methods described herein. The images 312 captured by the imaging apparatus 306 are two-dimensional images and therefore cell boundary profiles associated with the images 312 represent boundaries of the cells as a projection, or cross-sectional view, of the three-dimensional cells. The microfluidic arrangement 302, imaging apparatus 306 and processing of the images ensures that highly accurate relative measurements of images 312 can be made. Images 312 captured by the imaging apparatus 306 are streamed as image data to a computing device 308 directly, or indirectly via a network 310 comprising one or more computing devices.

[0041] Once the one or more images 312 have been captured at step SI 06, cell boundary profile data 314 is determined from the one or more images 312 at step S108. Following imaging of a cell subject to flow-induced deformation at step S106, one or more images 312 of cells are processed using any appropriate processing technique in order to extract cell boundary profile data 314. An example of extracting cell boundary profile data is described in greater detail with reference to Figures 5 A and 5B.

[0042] Figure 5A shows an exemplary method S500 of determining cell boundary profile data. The process is initiated at step S502 in response to one or more commands. For example, the process is initiated in response to obtaining one or more images 312 as described with reference to Figures 1 and 3. The process S500 is performed at the computing device 308 described with reference to Figure 3. In further examples, the process S500 is additionally or alternatively performed at one or more different computing devices. Once initiated at step S502, the process S500 moves to step S504, where one or more experimental images are captured. The experimental image relates to a cell that is in a region of interest. In an example, the region of interest is defined as a region associated with a section of the microfluidic channel, such as the microfluidic channel 304 of Figure 3, where a cell is subject to deformation conditions.

[0043] An example of an experimental image 520 of a cell is shown at Figure 5B. The experimental image 520 of a cell is correlated with data relating to the one or more deformation conditions applied to the cell. The experimental image 520 is a two-dimensional image of a cell taken by an imaging apparatus 306 as the cell flows through a microfluidic arrangement 302. In an example, the experimental image 520 is an image of the cell 303 subject to transient deformation. The one or more images 520 are optionally processed, for example to remove portions of the image relating to the microfluidic channel 304 and / or static noise.

[0044] Once the experimental image 520 of the cell has been obtained and processed at step S504, it is further processed at step S506, where the experimental image 520 is binarized. A binarized image 522 is shown at Figure 5B. The binarized image 522 is optionally processed to centre the image. Subsequently, the process S500 moves to step S508, where the cell boundary is extracted from the binarized image 522. An extracted cell boundary 524 is shown at Figure 5B. Optionally, the cell boundary is extracted by detecting the boundary with a border and sorting the boundary nodes in a counter-clockwise manner or a clockwise manner. The cell boundary is further extracted by fitting a cell boundary curve to the binarized data to provide cell boundary data corresponding to the fitted cell boundary curve. The cell boundary curve is fitted to the binarized data by applying a piecewise second-order polynomial function to the binarized data to provide cell boundary data corresponding to a fitted cell boundary curve.

[0045] Once the cell boundary has been extracted at step S508, the process S500 moves to step S510, where boundary nodes 526 are determined. Boundary nodes 526 are shown at Figure 5B. Boundary nodes 526 are determined by sampling data points with an equal arc-length separation function corresponding to the fitted cell boundary curve 524. An equal arc-length separation is the separation 528 between adjacent boundary nodes 526.

[0046] Once the boundary nodes 526 have been determined at step S510, the process S500 moves to step S512, where 1 -dimensional coordinates are determined in a 1-D coordinate vector 530. The process then ends at step S514.

[0047] Once cell boundary profile data has been extracted at step SI 08, for example using the process S500 described with reference to Figures 5A and 5B, the process S100 moves to step SI 10, wherein the cell boundary profile data, such as the cell boundary profile data 314 described with reference to Figure 3 is input into a machine learning model 316. In an example, the cell boundary profile data 314 is the 1-D coordinate vector 530 described with reference to Figure 5. In further examples, cell boundary profile data 314 is obtained through the processing of data in any appropriate manner in order to provide a suitable machine learning model input.

[0048] The machine learning model 316 is a model stored in the memory of the computing device 308. In further examples, the machine learning model 316 is stored on a different computing device, such as a computing device forming part of the network 310 of Figure 3.

[0049] The machine learning model 316 is an ultralow-latency Artificial-Intelligence-based model for accurately predicting, in real-time, multiple intrinsic mechanical properties of subcellular components of cells, based on the cell boundary profile data related to flow-induced deformation caused by flow of cells 303 through a microfluidic channel 304. In an example, the machine learning model 316 comprises at least one neural network. In an example the neural network is a multi-layer perceptron (MLP). In further examples, the machine learning model comprises a support vector machine (SVM), a random forest (RF) and / or a long short-term memory network (LSTM). In further examples, the machine learning model comprises any number of suitable additional and / or alternative architectures for computing in accordance with the methods described herein.

[0050] The machine learning model 316 is a model that has been trained using simulated boundary profile data corresponding to the steady and / or transient deformation of a cell subject to one or more steady and / or transient deformation conditions, wherein the simulated boundary profile data is based on a high-fidelity mechanistic model 318 of the cell comprising one or more intrinsic mechanical properties of one or more subcellular components of the cell. Advantageously, the machine learning model 316 is trained using labelled data from high- fidelity numerical simulations. Beneficially, the use of a mechanical model to provide simulations for training the machine learning model 316 avoids the need to obtain experimental data, which can be extremely challenging. In further examples, alternatively and / or additionally, the machine learning model 316 is trained using any appropriate techniques such that it is able to process cell boundary profile data in order to determine one or more intrinsic mechanical properties or size of one or more subcellular components of a cell, as described herein.

[0051] At Figure 2 there is shown a process flow S200 for generating training data and training a machine learning model, such as the machine learning model 316 described with reference to Figure 3.

[0052] The process S200 is initiated at step S202 in response to one or more commands. In order to train a machine learning model, training data is generated. If training data has already been generated, the process moves to step S208, where training data is input into a machine learning model. Optionally, in order to generate training data, the process moves to step S204, wherein a mechanistic cell model is constructed. At step S204, a mechanistic model of a cell is determined based on the intrinsic mechanical properties of one or more subcellular components of the cell. Figure 6A shows an example of a cell 600A based on a mechanistic model.Subcellular components of the cell 600A include a cell membrane 602, cell nucleus 604 and cytoplasm 606. The mechanistic model is constructed based on one or more intrinsic mechanical properties of the subcellular components, such as Young’s modulus, Poisson ratio and viscosity. Advantageously, a high-fidelity mechanical model is achieved by considering the cell as a three- layer structure with a viscoelastic membrane that represents the lipid bilayer and the underlying cell cortex, a viscous or viscoelastic cytoplasm, and a nucleus modelled as a smaller deformable capsule consisting of a viscoelastic nuclear envelop enclosing a viscous or viscoelastic nucleoplasm. In further examples, the mechanistic model is based on additional and / or alternative mechanical properties of subcellular components of a cell.

[0053] Once the mechanistic model has been established at step S204, the process S200 moves to step S206 where deformation of the cell is simulated. The deformation of the cell is simulated for a wide range of flow parameters, such as fluid medium rheology and flow speed and for different cell properties, such as size, shape and intrinsic mechanical properties of one or more subcellular components of the cell. Figure 6B shows exemplary simulations 600B for different parameters. In the example of Figure 6B, nine simulations of cell boundary profile data are shown based on three different cell sizes 606, 608, 610 and three different cell membrane elasticities 612, 614, 616. The simulated cell boundary profile data is correlated to the intrinsic mechanical properties of the subcellular components of the mechanistic cell model to provide training data. The generation of subsets of simulations for different cell deformations, for example those corresponding to transient and / or steady deformations of a cell, provides training data that is used to train a machine learning model such that the machine learning model can subsequently use cell boundary profile data as an input to determine the intrinsic mechanical properties and or size of one or more subcellular components of a cell.

[0054] An example of the use of the mechanistic model 600A as the basis for simulations of cell boundary profile data is described with reference to Figures 7A and 7B. Figures 7A and 7B show images of cells under different deformation conditions with corresponding simulated cell boundary profile data superimposed on the images.

[0055] Figure 7A shows a diagram 700A correlating images of real prostate cancer PC-3 cells subject to steady deformation with predicted cell deformation based on a mechanistic model of a cell. There are shown cells of 5 different sizes 702, 704, 706, 708, 710 for four different cell membrane elasticities 712, 714, 716, 718. The diagram 700A shows dashed boundary profiles closely tracking the boundaries of the cells in the images.

[0056] Figure 7B shows a series 700B of three transient deformations. A first transient deformation 720 relates to a prostate cancer PC-3 cell. The second transient deformation 722 relates to a leukaemia K-562 cell. The third transient deformation 724 relates to a freshlyisolated peripheral blood mononucleate cell. In each of the transient deformations there is shown a series of cell images and corresponding simulations of cell boundary profile data. The cells are shown to have been imaged at multiple times as the cells move in the direction of the arrow 726 through a respective microchannel. The cells are shown to be deformed differently during at different times and locations as they are subject to transient deformation. In each of the three cases, the simulated boundary profile data is shown closely to track the cell boundaries shown in the two-dimensional images.

[0057] The mechanistic model of the cells described with reference to Figures 7A and 7B is based on intrinsic mechanical properties of subcellular components of the cells. The close agreement between real measurements and simulations shows a mechanistic model that can be used to provide simulations for training a machine learning model, such as the machine learning model 316 described with reference to Figure 3.

[0058] Once sufficient training data has been obtained at step S206, the process moves to step S208, where the training data is input into a machine learning model, such as the machine learning model 316 described with reference to Figure 3. The machine learning model 316 trained in an iterative manner. The machine learning model 316 outputs a prediction at step S210, based on the validation data input at step S208. The process moves to step S212, where the prediction is compared to a ground truth and it is determined if the prediction is below a predetermined threshold. If it is not below a predetermined threshold, the process moves to step S214, where hyper-parameters of the machine learning algorithm are adjusted. The process then moves to step S208, where the adjusted model receives a further input and generates a further output at step S210. When it is determined at step S212 that the predicted output of the machine learning model is below a threshold, the process moves to step S216, where the trained machine learning model is determined. Beneficially, supervised training is used in order to determine a machine learning model that provides accurate output in response to receiving cell boundary profile data. Once the machine learning model has been determined, the process ends at step S218. Whilst the training and determination of a machine learning model is implemented at the computing device 308 of Figure 3, or at another computing device of the network 310 of Figure 3, in further examples, the machine learning model is trained and / or determined at any suitable computing device. Once a machine learning model 316 has been determined, cell boundary profile data can be input into the machine learning model 316.

[0059] For example, with reference to Figure 1, once the cell boundary profile data has been input into the machine learning model 316 at Step SI 10, the process SI 00 moves to step SI 12, where the cell boundary profile data 314 is processed by the machine learning model 316 inorder to determine one or more intrinsic mechanical properties and / or size of subcellular components of the cell are determined based on the cell boundary profile data.

[0060] The output of the machine learning model 316 takes any appropriate form. The output is optionally a direct output by the computing device 308, for example at a user interface of the computing device, 308, or to one or more computing devices in the network of one or more computing devices 310. In an example, the data is output in the form of plots, such as the plots 320, 322, 324 shown with reference to Figure 3.

[0061] Once the one or more intrinsic mechanical properties and / or size of one or more subcellular components of the cell have been determined at step SI 12, optionally the process SI 00 moves directly to the end step SI 16 and the determined one or more intrinsic mechanical properties and / or size of one or more subcellular components of the cell are output and / or stored, as appropriate.

[0062] Additionally, or alternatively, the machine learning model 316 determines further properties based on the determined one or more intrinsic mechanical properties or size of the subcellular components of the cell. Additionally, or alternatively, the output of the machine learning model 316 optionally serves as an input for further processing by one or more further algorithms to determine one or more further outputs based on the determined one or more intrinsic mechanical properties and / or size of the subcellular components of the cell. For example, optionally, once the one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell have been determined at step SI 12, the process moves to step SI 14, where the cell is classified and or the cell status is determined based on the determined one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell.

[0063] In an example, the cell status may be one of a cell in a healthy or a diseased state. Classification and / or status determination may result in an output at the computing device 308 and / or at a further computing device of the network 310 of one or more computing devices.

[0064] The classification of a cell type and / or cell status at step SI 14 of the process S100 is used to prognose and / or diagnose a condition of a subject and / or to monitor the efficacy of a drug used to treat the condition. For example, based on the determination of one or more intrinsic mechanical properties and / or size of the subcellular components of the one or more cells obtained from a sample of a subject, the cells can be correlated to a cell classification and / or cell status and serve as label-free biomarkers of a condition. In order to correlate cell classification and / or cell status to the intrinsic mechanical properties of the subcellular components of a cell, the multi-parameter distribution of cell properties for individual cell types is determined. For example, a comprehensive characterisation of intrinsic mechanical properties of a significantnumber of cells (e.g., thousands to many millions) is conducted in order to establish statistical distributions. The optimum criterion, using the intrinsic mechanical properties and sizes of the subcellular components for distinguishing cells of different types and statuses are then determined, for example, using a traversal algorithm. Such correlation enables the determination of conditions based on the one or more intrinsic mechanical properties and / or size of one or more subcellular components of one or more cells. In an example, the condition comprises cancer. In further examples, additionally or alternatively, the condition comprises one or more cardiovascular diseases, diabetes or sepsis. Further, the determination of the cell statuses of one or many cells from a sample from a subject can be used to determine any changes in status which may be indicative of the efficacy of a drug used to treat a condition. For example, repeating the process of obtaining a sample from a subject and determining the cell status using the methods described herein enables any changes between healthy and diseases statuses of cells to be monitored and correlated with the efficacy of any drug that a patient is receiving to treat a condition such as cancer, cardiovascular disease, diabetes and / or sepsis. In an example, cell status corresponds to at least one of a normal or healthy cell, a diseased cell, a hardened cell, a softened cell, an activated cell, an inactivated cell and a cell in a specific cell development cycle. Beneficially, such cell statuses can provide information that informs treatment and / or the efficacy of treatment, for example.

[0065] Once the cell has been classified and / or the cell status has been determined at step SI 14, the process SI 00 may move to step SI 18, where the process SI 00 ends.

[0066] Additionally, or alternatively, once one or more intrinsic mechanical properties and / or size of subcellular components of a cell have been determined at step SI 12, and / or once the classification of a cell type and / or a cell status have been determined at step SI 14, cells are optionally sorted at step SI 16, based on the outcome of steps SI 12 and / or S 114.

[0067] For example, at step SI 16, in response to classifying and / or determining the status of a cell, the cell can be sorted using a sorting arrangement in combination with the microfluidic arrangement 302 described with reference to Figure 3. In an example, the sorting arrangement is in fluid communication with an outlet of the microchannel 304 of the microfluidic arrangement 302. The sorting arrangement is configured selectively to direct cells in one of a plurality of directions based on the classified cell type and / or cell status. The sorting arrangement is a module that is configured to use a form of physical force to directly drive the cell in a certain direction, or to use a force to drive a fluid flow to indirectly drive the cell in a certain direction. In an example, the sorting arrangement generates a force by a piezoelectric component, electrostatic approach, optomechanical approach, the use of compressed air and / or valves.

[0068] Once cell sorting at step SI 16 has finished, the process S100 moves to step SI 18, where the process S100 ends. Whilst the steps of the process S100 are shown in a particular order, in further examples, the steps are performed in any appropriate order and / or simultaneously in order to make appropriate determinations. For example, whilst steps S104 and S106 are shown sequentially in the process flow SI 00 of Figure 1, it is understood that flow-induced deformation and high-speed imaging of a cell can be performed simultaneously.

[0069] In further examples, additionally, and / or alternatively steps of the process S100 and / or the entire process SI 00 are repeated in order to provide outputs.

[0070] Whilst steps S104 to S108 describe a method of imaging and determining cell boundary profile data 314, in further examples cell boundary profile data 314 is provided by any appropriate means as an input at step SI 10.

[0071] Beneficially, the machine learning model 316 enables fast and accurate determinations of intrinsic mechanical properties and / or size of subcellular components of a cell. For example, where the machine learning model 316 comprises a multi-layer perceptron, typical image processing and prediction times are 0.45 ms and 0.19 ms, respectively, with a prediction error of 6%. Where the machine learning model 316 comprises a support vector machine, typical image processing and prediction times are 0.45 ms and 0.43 ms, respectively, with a prediction error of 7%. Where the machine learning model 316 comprises a random forest, typical image processing and prediction times are 0.45 ms and 3.13 ms, respectively, with a prediction error of 6%. Where the machine learning model 316 comprises a long short term memory network, typical image processing and prediction times are 0.25 ms and 0.45 ms, respectively, with a prediction error of 3%. These times can be reduced by two orders of magnitude by optimising the algorithms and using advanced hardware such as field programmable gate arrays.

[0072] Examples of experimentally determined intrinsic mechanical properties of subcellular components of cells using a multi-layer perceptron model are described with reference to Figures 8A to 8B. Figures 8A to 8D show results or two different types of cells, obtained through the use of a multi-layer perceptron model trained using simulation data obtained from a mechanistic model based on subcellular components of a cell. Using high-throughput real-time imaging of cells subject to deformation, cell boundary profile data were extracted, as described herein with reference to Figure 1. Subsequently, cell boundary profile data were input into a multi-layer perceptron model and subcellular-level intrinsic mechanical properties were output for analysis.

[0073] Figure 8A shows a plot 800A of cell membrane elasticity on the y-axis 804A as a function of cell diameter on the x-axis 802A for prostate cancer PC3 cells with corresponding cumulative distributions for cell diameter 806A and cell membrane elasticity 808A. There are N=71,414 cells in the sample.

[0074] Figure 8B shows a plot 800B of cell membrane elasticity on the y-axis 804B as a function of cell diameter on the x-axis 802B for leukaemia K562 cells with corresponding cumulative distributions for cell diameter 806B and cell membrane elasticity 808B. There are N=96,998 cells in the sample.

[0075] Figure 8C shows plots 800C of a statistical distribution of frequency on the y-axis 810C as a function of cell diameter on the x-axis 802C for prostate cancer PC3 cells 806B and leukaemia K562 cells 806A based on data extracted from the plots 800A, 800B of Figures 8A and 8B, respectively.

[0076] Figure 8D shows plots 800D of a statistical distribution of frequency on the y-axis 810D as a function of cell membrane elasticity on the x-axis 802D for prostate cancer PC3 cells 808B and leukaemia K562 cells 808A based on data extracted from the plots 800A, 800B of Figures 8A and 8B, respectively. The same methodology of Figure 8 can also be used to characterise the membrane viscosity of cells of different types. Examples are shown in Figures 11 A and B, which plot the cell membrane viscosity as a function of cell diameter for PC-3 and K-562 cells respectively.

[0077] The results shown at Figures 8A to 8D and 11 A and B illustrate how the different cell types result in different distributions of subcellular-level cell intrinsic mechanical properties and cell sizes, providing evidence that cells can be classified based on determination of subcellular intrinsic mechanical properties and sizes.

[0078] Examples of intrinsic mechanical properties and sizes of subcellular components of cells determined using the present disclosed system for blood samples under different scenarios are described with reference to Figures 9A to 9C. In order to provide the outputs seen at Figures 9A to 9C, 27 different blood samples from three groups of patients were analysed: the first three samples associated with three healthy participants; the next ten samples associated with ten presurgery prostate cancer patients; and the final fourteen samples associated with fourteen postsurgery prostate cancer patients. The samples were subject the present system for flow-induced cell deformation in a straight channel in order to extract cell boundary profile data, as described herein. The cell boundary profile data were then inputted into a machine learning model that was trained based on simulations using a mechanistic cell model subject to different flow- induced deformation conditions, in order to output data regarding the intrinsic properties of the subcellular components.

[0079] Figures 9A shows a plot 900A of distribution of cell membrane elasticity along the y-axis 904A as a function of cell diameter along the x-axis 902A for freshly isolated peripheral blood mononucleate cells obtained from three healthy participants with corresponding cumulativedistributions for cell diameter 908A and cell membrane elasticity 910A. There are N=l, 704, 906 cells in the sample.

[0080] Figures 9B shows a plot 900B of distribution of cell membrane elasticity along the y-axis 904B as a function of cell diameter along the x-axis 902B for freshly isolated peripheral blood mononucleate cells obtained from ten pre-surgery prostate cancer patients with corresponding cumulative distributions for cell diameter 908B and cell membrane elasticity 910B. There are N=4, 161,546 cells in the sample.

[0081] Figures 9C shows a plot 900C of distribution of cell membrane elasticity along the y-axis 904C as a function of cell diameter along the x-axis 902C for freshly isolated peripheral blood mononucleate cells obtained from fourteen post-surgery prostate cancer patients with corresponding cumulative distributions for cell diameter 908C and cell membrane elasticity 910C. There are N=5,621,902 cells in the sample.

[0082] The distinctive features in the distributions of cell membrane elasticity vs cell diameter observed from blood samples from different patient groups strongly suggest that the distribution generated for blood samples using the present system can be used as a label-free biomarker for the determination of a condition such as prostate cancer, or the states of a condition such as pre- or post-surgery.

[0083] Alternatively, or in addition to determining intrinsic mechanical properties of subcellular components of a cell based on cell boundary profile data, the size of subcellular components of a cell can also be determined based on cell boundary profile data. Figure 10 shows simulated (curves) and experimental (dots) plots 1000 of the time evolutions of the cell deformation parameter, which is calculated from the cell boundary profile, when the cell is flowing through a cross-slot microchannel. Simulated results have covered cells with different sizes of nucleus. In particular, cell deformation parameter on the y-axis 1004 is plotted against time on the x-axis 1002 for three different ratios of cell nucleus size to cell size. The data points 1012 show experimental data. A first curve 1006 shows simulated results for a size ratio of 0.4. A second curve 1008 shows simulated results for a size ratio of 0.5. A third curve 1010 shows simulated results for a size ratio of 0.6. Excellent agreement between numerical simulation and experiment can only be achieved when the size ratio is set as 0.4 in the numerical simulation. Therefore, from the cell boundary profile one can infer the size of the cell nucleus, without directly visualising or touching the nucleus. In turn, such simulations can be used to train a machine learning model that is able to output the size of subcellular components of a cell based on the cell boundary profile data. The same methodology also enables one to predict the mechanical properties of the cell nucleus. Extensive experiments indicate that the overall prediction accuracy for cell nucleus size is 3%, and the accuracy for predicting cell nucleus mechanicalproperties such as the elasticity is 13%. Where the size and / or mechanical properties of the nucleus of the cell is indicative of a particular condition or status of a cell, such a determination can subsequently be used to classify the type and / or status of a cell based on cell boundary profile data.

[0084] The methods and steps described herein may be implemented at least partially by one or more computing devices. The methods and steps described herein may be stored on a computer readable medium in the form of instructions that, when executed by a computing device, perform at least some of the methods and steps described herein. Whilst examples of process and apparatuses are provided, it will be understood that in further examples there may be additional and / or alternative steps and / or apparatus in order to determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on cell boundary profile data, as set out herein.

Claims

CLAIMS:

1. A method of determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell, the method comprising: determining cell boundary profile data associated with a cell subject to steady or transient deformation; inputting the cell boundary profile data into a machine learning model; and using the machine learning model to determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of the cell based on the cell boundary profile data.

2. The method according to claim 1, wherein the method further comprises: classifying the cell type and / or determining the cell status based on the determined one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell.

3. The method according to any preceding claim, wherein determining cell boundary profile data associated with a cell subject to steady or transient deformation comprises: applying one or more steady and / or transient deformation conditions to cell; and capturing one or more images of the cell associated with the one or more steady and / or transient deformation conditions.

4. The method according to claim 3, wherein applying one or more steady and / or transient deformation conditions to the cell comprises: directing the cell to flow through one or more microchannels comprising a geometry configured to induce cell deformation, optionally wherein the one or more microchannels comprises a straight channel geometry, a converging channel geometry, a diverging channel geometry, a wavy channel geometry, a cross-slot channel geometry or a branched channel geometry, optionally wherein at least a portion of the channel has a cross-sectional profile that is square, rectangular, circular, semi-circular, elliptical or semi-elliptical.

5. The method according to claim 3 or claim 4, wherein determining cell boundary profile data comprises: determining one or more 1 -dimensional vectors of coordinates of cell boundary nodes based on the one or more images.

6. The method according to any of claims 3 to 5, wherein determining cell boundary profile data comprises: binarizing at least a portion of the one or more images to provide binarized data; extracting cell boundary data from the binarized data; determining nodes associated with the cell boundary data; and determining a coordinate vector based on the determined nodes.

7. The method according to claim 6, wherein extracting the cell boundary data comprises: fitting a cell boundary curve to the binarized data to provide cell boundary data corresponding to the fitted cell boundary curve, optionally wherein fitting a cell boundary curve to the binarized data comprises applying a piecewise second-order polynomial function to the binarized data to provide cell boundary data corresponding to a fitted cell boundary curve.

8. The method according to claim 7, wherein determining nodes associated with the cell boundary data comprises: sampling data points with an equal arc-length separation function corresponding to the fitted cell boundary curve.

9. The method according to any preceding claim, wherein the machine learning model has been trained using simulated cell boundary profile data corresponding to the steady and / or transient deformation of a cell subject to one or more steady and / or transient deformation conditions, wherein the simulated cell boundary profile data is based on a mechanistic model of the cell comprising one or more intrinsic mechanical properties of one or more subcellular components of the cell.

10. The method according to any of claims 2 to 9, the method further comprising: sorting the cell by selectively directing the cell in one of a plurality of directions based on the classified cell type and / or cell status by applying a physical force to directly drive the cell, or using a physical force to drive a fluid flow to direct the cell, optionally wherein thephysical force is generated by at least one of a piezoelectric component, an electrostatic approach, an optomechanical approach, by using compressed air, and / or valves.

11. The method according to any of claims 2 to 10, wherein the cell is obtained from a sample of a subject, the method further comprising: prognosing or diagnosing a condition of the subject and / or monitoring the efficacy of a drug based on the classified cell type and / or cell status of one or more cells.

12. The method according to claim 11, wherein the sample is a blood sample or the sample is a cell suspension, optionally wherein the cell suspension comprises cells obtained from a patient tissue by physically grinding or biochemically dissolving the patient tissue and suspending the physically ground or biochemically dissolved patient tissue in a liquid medium.

13. The method according to claim 11 or claim 12, wherein the condition comprises at least one of cancer, cardiovascular disease, diabetes and sepsis, optionally wherein the method comprises using the cell status to monitor the efficacy of drugs for treating the condition.

14. A method of training a machine learning model for determining one or more intrinsic mechanical properties and / or sizes of one or more subcellular components of a cell, the method comprising: receiving simulated cell boundary profile data corresponding to the steady and / or transient deformation of a cell subject to one or more steady and / or transient deformation conditions, wherein the simulated cell boundary profile data is based on a mechanistic model of the cell comprising one or more intrinsic mechanical properties of one or more subcellular components for a plurality of different parameters; and training the machine learning model to predict intrinsic mechanical properties and / or size of one or more subcellular components of a cell based on cell boundary profile of the cell.

15. The method of training a machine learning model according to claim 14, wherein training the machine learning model comprises supervised training.

16. The method according to any of claims 14 and 15, comprising generating simulated cell boundary profile data by:determining a mechanistic model of a cell based on the intrinsic mechanical properties of one or more subcellular components of the cell; using the mechanistic model to simulate cell boundary profile data associated with one or more steady and / or transient deformation conditions and / or cell properties, optionally wherein the one or more steady state and / or transient deformation conditions comprise fluid medium rheology and flow speed and / or wherein the cell properties comprise cell size, cell shape and / or intrinsic mechanical properties of one or more subcellular components of the cell.

17. The method according to any preceding claim, wherein the subcellular components include at least one of: a cell membrane, a cytoplasm and a nucleus.

18. The method according to any preceding claim, wherein the one or more intrinsic mechanical properties of the subcellular components comprise at least one of: a Young’s modulus, a Poisson ratio and a viscosity.

19. The method according to any preceding claim, wherein the machine learning model comprises at least one neural network, optionally wherein the neural network is a multi-layer perceptron, a support vector machine, a random forest or a long short-term memory network.

20. The method according to any of claims 2 to 19, wherein the cell type corresponds to a normal human cell type or a diseased cell type, optionally wherein the diseased cell type is a cancerous cell type, optionally wherein the cell type is a prostate cancer cell type or a leukaemia cell type.

21. The method according to any of claims 2 to 20, wherein the cell status corresponds to a normal / healthy, or a hardened / softened, or an activated / inactivated, or a cell in a specific cell development cycle.

22. A computing device configured to perform the method of any of claims 1 to 21.

23. A computer readable medium comprising instructions that, when executed by a computer, cause the computer to perform the method of any of claims 1 to 21.

24. A system for cell analysis, the system comprising:a microfluidic arrangement for controllably subjecting one or more cells to steady and / or transient deformation conditions; an imaging apparatus for imaging the one or more cells subject to steady and / or transient deformation conditions; and a computing device configured to perform the method of any of claims 1 to 9 and 11 to 21.

25. A system for cell sorting, the system comprising: a microfluidic arrangement for controllably subjecting one or more cells to steady and / or transient deformation conditions; an imaging apparatus for imaging the one or more cells subject to steady and / or transient deformation conditions; a computing device configured to perform the method of any of claim 10 and claims 11 to 21 when dependent on claim 10; and a cell sorting arrangement for selectively directing the one or more cells in one of a plurality of directions26. A machine learning model for determining one or more intrinsic mechanical properties and / or size of one or more subcellular components of a cell, wherein the machine learning model is configured to: receive cell boundary profile data corresponding to a steadily and / or transiently deformed cell; and determine one or more intrinsic mechanical properties and / or size of one or more subcellular components of the transiently deformed cell based on cell boundary profile data.

27. The machine learning model according to claim 26, wherein the machine learning model is further configured to: classify the cell type and / or cell status based on the determined one or more intrinsic mechanical properties and / or size of the one or more subcellular components of the cell.

28. The machine learning model according to claim 27, wherein the machine learning model comprises at least one neural network, optionally wherein the neural network is a multi-layer perceptron, a support vector machine, a random forest or a long short-term memory network.

Images