Towards Self-explainable Transformers for Cell Classification in Flow Cytometry Data

Simple SummaryB-cell Acute Lymphoblastic Leukaemia is one of the most common cancers in childhood, with 20% of patients eventually relapsing. Flow cytometry is routinely used for diagnosis and follow-up, but it currently does not provide prognostic value at diagnosis. The volume and the high-dimensional character of this data makes it ideal for its exploitation by means of Artificial Intelligence methods. We collected flow cytometry data from 56 patients from two hospitals. We analysed differences in intensity of marker expression in order to predict relapse at the moment of diagnosis. We finally correlated this data with biomolecular information, constructing a classifier based on CD38 expression.Artificial intelligence methods may help in unveiling information that is hidden in high-dimensional oncological data. Flow cytometry studies of haematological malignancies provide quantitative data with the potential to be used for the construction of response biomarkers. Many computational methods from the bioinformatics toolbox can be applied to these data, but they have not been exploited in their full potential in leukaemias, specifically for the case of childhood B-cell Acute Lymphoblastic Leukaemia. In this paper, we analysed flow cytometry data that were obtained at diagnosis from 56 paediatric B-cell Acute Lymphoblastic Leukaemia patients from two local institutions. Our aim was to assess the prognostic potential of immunophenotypical marker expression intensity. We constructed classifiers that are based on the Fisher’s Ratio to quantify differences between patients with relapsing and non-relapsing disease. We also correlated this with genetic information. The main result that arises from the data was the association between subexpression of marker CD38 and the probability of relapse.

Flow cytometry is a technique for analyzing cells that are suspended in a buffered salt-based solution and flow past one or more lasers. Visible light scatter and one or more fluorescence characteristics are assessed for each particle. The most common use of flow cytometry is immunophenotyping. In traditional flow cytometry analysis, a region around a population of cells is manually drawn (gating) in two-dimensional scatter plots. This enables the measuring of specific groups of cells. However, manual gating could increase the overall error rate of the study and makes the analysis hardly reproducible since it is done in less controlled settings. Furthermore, it represents a bottleneck in the analysis of large amounts of data. Some processes, such as gating, have recently been automated using R packages. In addition, dimensionality reduction approaches have been developed in the flow cytometry environment to take advantage of information from several markers at once. Despite this, none of them are integrated in a harmonized way, and none of them allow back-gating to emphasize the study population and eliminate false positives. To tackle this problem, our team has developed flowTOTAL (github.com/ImmunoOncology/flowTOTAL), a user-friendly command line workflow to analyze flow cytometry data. The major attractive feature is the facility to perform with one command not only a traditional analysis, but also an unsupervised analysis. As input, the user has to indicate the folder with the .FCS files, the metadata associated with each file, and the marker to be used during back-gating. The pipeline is divided into three main sections: preprocessing, traditional analysis, and unsupervised analysis. During preprocessing each. FCS will be subjected to correcting for fluorescence spillover (compensation), detection of anomalies by checking flow rate and signal acquisition as well as removing doublets based on forward scatter (QC). For the traditional analysis, auto-gating will be performed for the identification of the target population using back-gating. For each set of given markers, the number of events obtained and the scatter plot will be generated. Finally, in the unsupervised analysis, the population of interest will be specified and the pipeline will proceed with normalization, dimensionality reduction using PCA or UMAP and finally a clustering approach for subpopulation identification. In addition, differential abundance analysis can be performed with metadata information. flowTOTAL is presented as a standardization for the analysis of flow cytometry data, comprising all the necessary steps for comprehensive analysis and allowing mass analysis. Furthermore, it goes beyond the simple quantification of particles, since the implementation of more complex methodologies allows for the discovery of subpopulations that are not present in the traditional analysis but have a significant biological role. Citation Format: Juan Luis Onieva, Patricia Cháves, Javier Oliver, María Garrido-Barros, Juan Zafra, Belén Sojo, Alfonso Sánchez, Martina Álvarez, Pedro Jiménez, Emilio Alba, Miguel Berciano, Antonio Rueda, Manuel Cobo-Dols, Elisabeth Pérez, Isabel Barragán. flowTOTAL: A comprehensive bioinformatics workflow for flow cytometry automatic analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1910.

Automated clustering workflows are increasingly used for the analysis of high parameter flow cytometry data. This trend calls for algorithms which are able to quickly process tens of millions of data points, to compare results across subjects or time points, and to provide easily actionable interpretations of the results. To this end, we created Tailor, a model-based clustering algorithm specialized for flow cytometry data. Our approach leverages a phenotype-aware binning scheme to provide a coarse model of the data, which is then refined using a multivariate Gaussian mixture model. We benchmark Tailor using a simulation study and two flow cytometry data sets, and show that the results are robust to moderate departures from normality and inter-sample variation. Moreover, Tailor provides automated, non-overlapping annotations of its clusters, which facilitates interpretation of results and downstream analysis. Tailor is released as an R package, and the source code is publicly available at www.github.com/matei-ionita/Tailor.

The maternal-fetal interface represents a unique immune privileged site that maintains the ability to defend against pathogens while orchestrating the necessary tissue remodeling required for placentation. The recent discovery of novel cellular families (innate lymphoid cells, tissue-resident NK cells) suggests that our understanding of the decidual immunome is incomplete. To understand this complex milieu, new technological developments allow reproductive immunologists to collect increasingly complex data at a cellular resolution. Polychromatic flow cytometry allows for greater resolution in the identification of novel cell types by surface and intracellular protein. Single-cell RNA-seq coupled with microfluidics allows for efficient cellular transcriptomics. The extreme dimensionality and size of data sets generated, however, requires the application of novel computational approaches for unbiased analysis. There are now multiple dimensionality reduction (tSNE, SPADE) and visualization tools (SPICE) that allow researchers to efficiently analyze flow cytometry data. Development of computational tools has also been extended to RNA-seq data (including scRNA-seq), which requires specific analytical tools. Here, we provide an overview and a brief primer for the reproductive immunology community on data acquisition and computational tools for the analysis of complex flow cytometry and RNA-seq data.

A rule-based expert system is being developed to assist clinicians in the analysis of multivariate flow cytometry data for patients with leukemias or lymphomas. The cells are stained with fluorescently labeled monoclonal antibodies and the cell fluorescence is measured with a flow cytometer. Cluster analysis is used to isolate subpopulations in the data on which the clinical decisions are made. Symbolic facts for the expert system are instantiated using these numerical data and the knowledge of the clinicians and experts in flow cytometry. The first prototype used a decision tree and rigid rules. Is successfully classified only nine of eleven leukemia cases. A second prototype incorporating certainty factors into the rules is now being developed that should remove the need for a rigid decision tree. 9 refs.

BCyto: A shiny app for flow cytometry data analysis

Multiple myeloma, the second most common hematologic malignancy, is characterized by recurrent molecular genetic variants which portend important prognostic and therapeutic implications. These variants, which include chromosomal aneuploidies, translocations, and other structural changes, are currently identified by performing karyotyping and fluorescence in situ hybridization analysis on bone marrow biopsy samples. Comprehensive variant testing for full characterization of multiple myeloma cases can take days to weeks, is costly, and is not available to all patients. Past studies have explored to what extent cellular morphologic information from bone marrow aspirates or immunophenotypic information from flow cytometry can predict these genetic variants via the use of machine learning algorithms. However, the integration of multiple hematopathology data modalities using machine learning towards improving the diagnosis and characterization of hematologic neoplasms has yet to be explored. To this end, we developed a deep learning-based algorithm which, for individual patients, integrates tens-of-thousands of cell images from Wright-stained bone marrow aspirate smears with hundreds-of-thousands of events from myeloma flow cytometry panels to predict the presence or absence of particular molecular genetic variants among 85 cases of multiple myeloma. Attention-based multi-instance learning models were developed utilizing convolutional neural networks to analyze cell image data and multi-layer perceptrons to analyze flow cytometry data. Models trained solely on cell image data showed variable performance in predicting genetic subtypes of multiple myeloma cases: t(11;14), AUROC = 0.795; 1q+, AUROC = 0.743; del(17p), AUROC = 0.520. Similar models trained solely on flow cytometry data showed fair performance for all subtypes: t(11;14), AUROC = 0.761; 1q+, AUROC = 0.761; del(17p), AUROC = 0.787. However, models which combine features among high-attention events from both cell image and flow cytometry data yield significantly improved performance: t(11;14), AUROC = 0.892; 1q+, AUROC = 0.789; del(17p), AUROC = 0.876. Similarly, models trained to predict established prognostic classifications of multiple myeloma cases performed best when cell image and flow cytometry data is integrated: cell images alone, AUROC = 0.721; flow cytometry alone, AUROC = 0.827; both data combined, AUROC = 0.839. Additional improvements in prognostic classification were obtained when utilizing a Kronecker product to more completely utilize pairwise interactions between cell image and flow cytometry features: AUROC = 0.864. This study is, to our knowledge, the first to utilize deep learning to integrate multiple hematopathology data modalities towards automated diagnosis or subclassification of hematologic neoplasms in individual patients. The results of this study demonstrate the feasibility of accurately predicting molecular genetic subtypes of multiple myeloma cases solely from cell morphology and flow cytometric information. With further refinement, these promising models could be applied in low resource settings as well as research settings to better understand the biological relationship between plasma cell genetics with morphology and immunophenotype.

This unit provides several approaches to flow cytometry data analysis. Frequency determinations based on analysis of single-parameter fluorescence histograms and dual-parameter contour plots are presented. Next, steps are described for calculating values for signal-to-noise ratios when logarithmic amplification is used for data collection. This calculation can be used to compare the amounts of antigenic determinants per cell between different cells stained with the same reagent. Finally, a procedure is described for constructing a calibration curve for logarithmic amplifiers. This calibration curve is required for calculation of signal-to-noise values and can also be used to determine if the amplifiers are working properly.

Summary form only given. A description is given of the application of neural net computer technology to the analysis of flow cytometry data. Although the data used in this study are from oceanographic research, the results are general and should be directly applicable to flow cytometry data or any sort. The neural network described offers the advantages of adaptability to changing conditions and potential real-time analysis. High accuracy and processing speed, near that required for real-time classification, have been achieved in a software simulation of the neural network on a Macintosh SE personal computer. >

T-lineage acute lymphoblastic leukemia (T-ALL) accounts for about 15% of pediatric and about 25% of adult ALL cases. Minimal/measurable residual disease (MRD) assessed by flow cytometry (FCM) is an important prognostic indicator for risk stratification. In order to assess the MRD a limited number of antibodies directed against the most discriminative antigens must be selected. We propose a pipeline for evaluating the influence of different markers for cell population classification in FCM data. We use linear support vector machine, fitted to each sample individually to avoid issues with patient and laboratory variations. The best separating hyperplane direction as well as the influence of omitting specific markers is considered. Ninety-one bone marrow samples of 43 pediatric T-ALL patients from five reference laboratories were analyzed by FCM regarding marker importance for blast cell identification using combinations of eight different markers. For all laboratories, CD48 and CD99 were among the top three markers with strongest contribution to the optimal hyperplane, measured by median separating hyperplane coefficient size for all samples per center and time point (diagnosis, Day 15, Day 33). Based on the available limited set tested (CD3, CD4, CD5, CD7, CD8, CD45, CD48, CD99), our findings prove that CD48 and CD99 are useful markers for MRD monitoring in T-ALL. The proposed pipeline can be applied for evaluation of other marker combinations in the future.

Allogeneic Hematopoietic Stem Cell Transplantation Overcomes the Adverse Prognostic Impact of CD20 Expression In Both Adult and Pediatric Acute Lymphoblastic Leukemia.

In immunofluorescence measurements and most other flow cytometry applications, fluorescence signals of interest can range down to essentially zero. After fluorescence compensation, some cell populations will have low means and include events with negative data values. Logarithmic presentation has been very useful in providing informative displays of wide-ranging flow cytometry data, but it fails to adequately display cell populations with low means and high variances and, in particular, offers no way to include negative data values. This has led to a great deal of difficulty in interpreting and understanding flow cytometry data, has often resulted in incorrect delineation of cell populations, and has led many people to question the correctness of compensation computations that were, in fact, correct. We identified a set of criteria for creating data visualization methods that accommodate the scaling difficulties presented by flow cytometry data. On the basis of these, we developed a new data visualization method that provides important advantages over linear or logarithmic scaling for display of flow cytometry data, a scaling we refer to as "Logicle" scaling. Logicle functions represent a particular generalization of the hyperbolic sine function with one more adjustable parameter than linear or logarithmic functions. Finally, we developed methods for objectively and automatically selecting an appropriate value for this parameter. The Logicle display method provides more complete, appropriate, and readily interpretable representations of data that includes populations with low-to-zero means, including distributions resulting from fluorescence compensation procedures, than can be produced using either logarithmic or linear displays. The method includes a specific algorithm for evaluating actual data distributions and deriving parameters of the Logicle scaling function appropriate for optimal display of that data. It is critical to note that Logicle visualization does not change the data values or the descriptive statistics computed from them.

Death receptor 3 (DR3) is a tumor necrosis factor receptor superfamily member (TNFRSF25), which is minimally expressed on resting conventional T cells (though readily inducible upon cell activation), yet highly expressed on resting FoxP3+ regulatory T cells (Treg). We recently demonstrated that activation of DR3 with an agonistic antibody (4C12) leads to selective expansion and activation of Treg in healthy mice and suppression of graft-versus-host disease (GVHD) in recipient mice when donor mice are treated. However, given the long antibody half-life and concomitant safety concerns, along with the lack of a humanized agonistic antibody to DR3, both human and murine fusion proteins incorporating the natural DR3 ligand TL1A (TL1A-Ig) have been developed. Herein, we show that DR3 activation with 4C12 or with TL1A-Ig, with or without the addition of low dose IL-2 to the treatment regimen, led to a significant expansion of murine Treg in spleen, lymph nodes, and peripheral blood. Bioluminescent imaging revealed peak Treg expansion around day 7–8, with return to near baseline after 2–3 weeks. In addition to expansion, all DR3 agonist treatment regimens led to increased activation of Tregs, with significant upregulation of the activation markers ICOS, KLRG-1, PD-1, and CD103, and the proliferation marker Ki-67. The near absence of activated Treg populations in control treated spleens was also detected on tSNE analysis of flow cytometry data. Subtly different patterns of splenic Treg activation by the different DR3 agonists were noted in both tSNE analysis of flow cytometry data and RNA-sequencing analysis. However, upregulation of gene transcripts which play important roles in cell proliferation, trafficking, activation, and effector function were observed regardless of the DR3 agonist treatment regimen used. In the major MHC-mismatch model of hematopoietic cell transplantation, DR3 agonist-mediated expansion and activation of Tregs in donor mice led to a significant improvement in GVHD in recipient mice. These data provide important preclinical information regarding the outcome of DR3 activation with an agonistic antibody or natural ligand and provide insight into the therapeutic use of this approach to reduce GVHD in recipients and improve outcomes of hematopoietic cell transplantation.

Standardization, data mining techniques, and comparison to normality are changing the landscape of multiparameter flow cytometry in clinical hematology. On the basis of these principles, a strategy was developed for measurable residual disease (MRD) assessment. Herein, suspicious cell clusters are first identified at diagnosis using a clustering algorithm. Subsequently, automated multidimensional spaces, named “Clouds”, are created around these clusters on the basis of density calculations. This step identifies the immunophenotypic pattern of the suspicious cell clusters. Thereafter, using reference samples, the “Abnormality Ratio” (AR) of each Cloud is calculated, and major malignant Clouds are retained, known as “Leukemic Clouds” (L-Clouds). In follow-up samples, MRD is identified when more cells fall into a patient’s L-Cloud compared to reference samples (AR concept). This workflow was applied on simulated data and real-life leukemia flow cytometry data. On simulated data, strong patient-dependent positive correlation (R2 = 1) was observed between the AR and spiked-in leukemia cells. On real patient data, AR kinetics was in line with the clinical evolution for five out of six patients. In conclusion, we present a convenient flow cytometry data analysis approach for the follow-up of hematological malignancies. Further evaluation and validation on more patient samples and different flow cytometry panels is required before implementation in clinical practice.

Background. The sodium-dependent phosphate transporter NaPi2b is expressed in a number of tumors and serves as a target for monoclonal antibody therapy. The NaPi2b transporter has a large extracellular domain containing 4 cysteines and an MX35 epitope recognized by the corresponding monoclonal antibodies. Aim. To study the recognition of the MX35 epitope in cysteine mutants C303 and C350 of the sodium-dependent phosphate transporter NaPi2b by monoclonal antibodies in ovarian cancer cells. Material and methods. Ovarian cancer cells of the OVCAR-8 line were transfected with plasmids encoding mutant forms of the NaPi2b transporter. The recognition of these forms by monoclonal antibodies was studied by Western blot and flow cytometry. Statistical analysis of flow cytometry data was performed using Pearsons Chi-squared test with Yates correction. Results. Western blot analysis of lysates of OVCAR-8 cells expressing the wild-type NaPi2b transporter demonstrated the presence of a specific signal only in samples without the addition of dithiothreitol, while in all samples with NaPi2b cysteine mutants, even without the addition of dithiothreitol, no specific signal from monoclonal antibodies was detected. The flow cytometry results showed that proportion of OVCAR-8 cells with positive staining of the MX35 epitope with antibodies is less when expressing the NaPi2b mutants compared to the wild type. Conclusion. Cysteines C303 and C350 are involved in disulfide bonds formation, which is of key importance for the formation of the conformation of the extracellular domain of the sodium-dependent phosphate transporter NaPi2b, which ensures the MX35 epitope recognition by monoclonal antibodies.