UMAP Based Anomaly Detection for Minimal Residual Disease Quantification within Acute Myeloid Leukemia

Multiplex immunophenotyping of human acute myeloid leukemia patients revealed single -cell heterogeneity with special attention on therapy sensitive and therapy resistant subpopulations

Introduction

Understanding leukemia-associated immunophenotypes (LAIP) could assist in the design of therapies to ameliorate patient benefits in acute myeloid leukemia (AML). In our study, focusing on single-cell heterogeneity in therapeutic resistance, flow cytometric immunophenotyping of the peripheral blood of therapy-naive and follow-up AML patients versus age and sex-matched healthy controls (HCs) was performed.

Methods

The FACS panel consisted of Viobility 405/520 Fixable Dye, Anti-human CD45, CD19, CD3, CD7, CD33, CD34, CD38, CD64, CD117, CD135, HLA-DR antibodies. Unsupervised clustering algorithms such as Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) and Flow cytometry data that builds Self-Organizing Maps (FlowSOM) were used to reveal the LAIP. The measurable residual disease (MRD) was monitored by our proposed manual gating. To complement the characterization of peripheral immune cells, Luminex MAGPIX was used to measure the concentration of 31 soluble immune-oncology mediators from the plasma of AML patients and HC.

Results

Both manual gating, UMAP and FlowSOM showed normalization of LAIP similar to the HC immune landscape following therapy. Eleven metaclusters (MCs) were associated with AML before therapy. The follow-up of AML samples revealed four MCs of therapy sensitive cells, and one MC composed of therapeutic resistant cells (MC12: CD3-CD7-CD33-CD38- CD64- HLA-DR- CD117- CD135-) identified by the FlowSOM analysis. The initial AML blasts in the MRD gate (CD19-, CD45+, CD3-, CD38+/CD34±, CD7+/CD117+, CD117+/CD135+) were detectable at the lowest frequency in our current study at 22 cells per 100,000 (0.022%) CD45+CD3- living singlet parental population. In the plasma of AML patients the levels of BAFF, B7-H2, B7-H4, CD25, MICA, and Siglec-7 were increased versus HCs.

Conclusions

This study focused on understanding the LAIP in AML before and after therapeutic intervention. The study highlights the potential of using single-cell LAIP profiling and immune mediator measurements to monitor therapy response and identify measurable residual disease and therapy resistant cell populations in AML.

Unraveling genotype-phenotype associations and predictive modeling of outcome in acute myeloid leukemia

Acute myeloid leukemia (AML) comprises 32% of adult leukemia cases, with a 5-year survival rate of only 20-30%. Here, the immunophenotypic landscape of this heterogeneous malignancy is explored in a single-center cohort using a novel quantitative computational pipeline. For 122 patients who underwent induction treatment with intensive chemotherapy, leukemic cells were identified at diagnosis, computationally preprocessed, and quantitatively subtyped. Computational analysis provided a broad characterization of inter- and intra-patient heterogeneity, which would have been harder to achieve with manual bivariate gating. Statistical testing discovered associations between CD34, CD117, and HLA-DR expression patterns and genetic abnormalities. We found the presence of CD34+ cell populations at diagnosis to be associated with a shorter time to relapse. Moreover, CD34- CD117+ cell populations were associated with a longer time to AML-related mortality. Machine learning (ML) models were developed to predict 2-year survival, European LeukemiaNet (ELN) risk category, and inv(16) or NPM1mut, based on computationally quantified leukemic cell populations and limited clinical data, both readily available at diagnosis. We used explainable artificial intelligence (AI) to identify the key clinical characteristics and leukemic cell populations important for our ML models when making these predictions. Our findings highlight the importance of developing objective computational pipelines integrating immunophenotypic and genetic information in the risk stratification of AML.

Computational assessment of measurable residual disease in acute myeloid leukemia using mixture models

BACKGROUND

The proportion of residual leukemic blasts after chemotherapy assessed by multiparameter flow cytometry, is an important prognostic factor for the risk of relapse and overall survival in acute myeloid leukemia (AML). This measurable residual disease (MRD) is used in clinical trials to stratify patients for more or less intensive consolidation therapy. However, an objective and reproducible analysis method to assess MRD status from flow cytometry data is lacking, yet is highly anticipated for broader implementation of MRD testing.

METHODS

We propose a computational pipeline based on Gaussian mixture modeling that allows a fully automated assessment of MRD status while remaining completely interpretable for clinical diagnostic experts. Our pipeline requires limited training data, which makes it easily transferable to other medical centers and cytometry platforms.

RESULTS

We identify all healthy and leukemic immature myeloid cells in with high concordance (Spearman's Rho = 0.974) and classification performance (median F-score = 0.861) compared to manual analysis. Using control samples (n = 18), we calculate a computational MRD percentage with high concordance to expert gating (Spearman's rho = 0.823) and predict MRD status in a cohort of 35 AML follow-up measurements with high accuracy (97%).

CONCLUSIONS

We demonstrate that our pipeline provides a powerful tool for fast (~3 s) and objective automated MRD assessment in AML.

Using UMAP for Partially Synthetic Healthcare Tabular Data Generation and Validation

In healthcare, vast amounts of data are increasingly collected through sensors for smart health applications and patient monitoring or diagnosis. However, such medical data often comprise sensitive patient information, posing challenges regarding data privacy, and are resource-intensive to acquire for significant research purposes. In addition, the common case of lack of information due to technical issues, transcript errors, or differences between descriptors considered in different health centers leads to the need for data imputation and partial data generation techniques. This study introduces a novel methodology for partially synthetic tabular data generation, designed to reduce the reliance on sensor measurements and ensure secure data exchange. Using the UMAP (Uniform Manifold Approximation and Projection) visualization algorithm to transform the original, high-dimensional reference data set into a reduced-dimensional space, we generate and validate synthetic values for incomplete data sets. This approach mitigates the need for extensive sensor readings while addressing data privacy concerns by generating realistic synthetic samples. The proposed method is validated on prostate and breast cancer data sets, showing its effectiveness in completing and augmenting incomplete data sets using fully available references. Furthermore, our results demonstrate superior performance in comparison to state-of-the-art imputation techniques. This work makes a dual contribution by not only proposing an innovative method for synthetic data generation, but also studying and establishing a formal framework to understand and solve synthetic data generation and imputation problems in sensor-driven environments.

Automation in Flow Cytometry

Automation in clinical flow cytometry has the potential to revolutionize the field by improving processes and enhancing efficiency and accuracy. Integrating advanced robotics and artificial intelligence, these technologies can streamline sample preparation, data acquisition, and analysis. Automated sample handling reduces human error and increases throughput, allowing laboratories to handle larger volumes with consistent precision. Intelligent algorithms contribute to rapid data interpretation, aiding in the identification of cellular markers for disease diagnosis and monitoring. This automation not only accelerates turnaround times but also ensures reproducibility, making clinical flow cytometry a reliable tool in the realm of personalized medicine and diagnostics.

MAGIC-DR: An interpretable machine-learning guided approach for acute myeloid leukemia measurable residual disease analysis

Multiparameter flow cytometry is widely used for acute myeloid leukemia minimal residual disease testing (AML MRD) but is time consuming and demands substantial expertise. Machine learning offers potential advancements in accuracy and efficiency, but has yet to be widely adopted for this application. To explore this, we trained single cell XGBoost classifiers from 98 diagnostic AML cell populations and 30 MRD negative samples. Performance was assessed by cross-validation. Predictions were integrated with UMAP as a heatmap parameter for an augmented/interactive AML MRD analysis framework, which was benchmarked against traditional MRD analysis for 25 test cases. The results showed that XGBoost achieved a median AUC of 0.97, effectively distinguishing diverse AML cell populations from normal cells. When integrated with UMAP, the classifiers highlighted MRD populations against the background of normal events. Our pipeline, MAGIC-DR, incorporated classifier predictions and UMAP into flow cytometry standard (FCS) files. This enabled a human-in-the-loop machine learning guided MRD workflow. Validation against conventional analysis for 25 MRD samples showed 100% concordance in myeloid blast detection, with MAGIC-DR also identifying several immature monocytic populations not readily found by conventional analysis. In conclusion, Integrating a supervised classifier with unsupervised dimension reduction offers a robust method for AML MRD analysis that can be seamlessly integrated into conventional workflows. Our approach can support and augment human analysis by highlighting abnormal populations that can be gated on for quantification and further assessment. This has the potential to speed up MRD analysis, and potentially improve detection sensitivity for certain AML immunophenotypes.

Identifying Novel Subtypes of Functional Gastrointestinal Disorder by Analyzing Nonlinear Structure in Integrative Biopsychosocial Questionnaire Data

Background/Objectives: Given the limited success in treating functional gastrointestinal disorders (FGIDs) through conventional methods, there is a pressing need for tailored treatments that account for the heterogeneity and biopsychosocial factors associated with FGIDs. Here, we considered the potential of novel subtypes of FGIDs based on biopsychosocial information. Methods: We collected data from 198 FGID patients utilizing an integrative approach that included the traditional Korean medicine diagnosis questionnaire for digestive symptoms (KM), as well as the 36-item Short Form Health Survey (SF-36), alongside the conventional Rome-criteria-based Korean Bowel Disease Questionnaire (K-BDQ). Multivariate analyses were conducted to assess whether KM or SF-36 provided additional information beyond the K-BDQ and its statistical relevance to symptom severity. Questions related to symptom severity were selected using an extremely randomized trees (ERT) regressor to develop an integrative questionnaire. For the identification of novel subtypes, Uniform Manifold Approximation and Projection and spectral clustering were used for nonlinear dimensionality reduction and clustering, respectively. The validity of the clusters was assessed using certain metrics, such as trustworthiness, silhouette coefficient, and accordance rate. An ERT classifier was employed to further validate the clustered result. Results: The multivariate analyses revealed that SF-36 and KM supplemented the psychosocial aspects lacking in K-BDQ. Through the application of nonlinear clustering using the integrative questionnaire data, four subtypes of FGID were identified: mild, severe, mind-symptom predominance, and body-symptom predominance. Conclusions: The identification of these subtypes offers a framework for personalized treatment strategies, thus potentially enhancing therapeutic outcomes by tailoring interventions to the unique biopsychosocial profiles of FGID patients.

Preterm infants show an atypical processing of the mother's voice

To understand the consequences of prematurity on language perception, it is fundamental to determine how atypical early sensory experience affects brain development. At term equivalent age, ten preterm and ten full-term newborns underwent high-density EEG during mother or stranger speech presentation, in the forward or backward order. A general group effect terms > preterms is evident in the theta frequency band, in the left temporal area, with preterms showing significant activation for strangers' and terms for the mother's voice. A significant group contrast in the low and high theta in the right temporal regions indicates higher activations for the stranger's voice in preterms. Finally, only full terms presented a late gamma band increase for the maternal voice, indicating a more mature brain response. EEG time-frequency analysis demonstrate that preterm infants are selectively responsive to stranger voices in both temporal hemispheres, and that they lack selective brain responses to their mother's forward voice.

Artificial Intelligence Enhances Diagnostic Flow Cytometry Workflow in the Detection of Minimal Residual Disease of Chronic Lymphocytic Leukemia

Flow cytometric (FC) immunophenotyping is critical but time-consuming in diagnosing minimal residual disease (MRD). We evaluated whether human-in-the-loop artificial intelligence (AI) could improve the efficiency of clinical laboratories in detecting MRD in chronic lymphocytic leukemia (CLL). We developed deep neural networks (DNN) that were trained on a 10-color CLL MRD panel from treated CLL patients, including DNN trained on the full cohort of 202 patients (F-DNN) and DNN trained on 138 patients with low-event cases (MRD < 1000 events) (L-DNN). A hybrid DNN approach was utilized, with F-DNN and L-DNN applied sequentially to cases. “Ground truth” classification of CLL MRD was confirmed by expert analysis. The hybrid DNN approach demonstrated an overall accuracy of 97.1% (95% CI: 84.7−99.9%) in an independent cohort of 34 unknown samples. When CLL cells were reported as a percentage of total white blood cells, there was excellent correlation between the DNN and expert analysis [r > 0.999; Passing−Bablok slope = 0.997 (95% CI: 0.988−0.999) and intercept = 0.001 (95% CI: 0.000−0.001)]. Gating time was dramatically reduced to 12 s/case by DNN from 15 min/case by the manual process. The proposed DNN demonstrated high accuracy in CLL MRD detection and significantly improved workflow efficiency. Additional clinical validation is needed before it can be fully integrated into the existing clinical laboratory practice.

Fuzzy Information Discrimination Measures and Their Application to Low Dimensional Embedding Construction in the UMAP Algorithm

Dimensionality reduction techniques are often used by researchers in order to make high dimensional data easier to interpret visually, as data visualization is only possible in low dimensional spaces. Recent research in nonlinear dimensionality reduction introduced many effective algorithms, including t-distributed stochastic neighbor embedding (t-SNE), uniform manifold approximation and projection (UMAP), dimensionality reduction technique based on triplet constraints (TriMAP), and pairwise controlled manifold approximation (PaCMAP), aimed to preserve both the local and global structure of high dimensional data while reducing the dimensionality. The UMAP algorithm has found its application in bioinformatics, genetics, genomics, and has been widely used to improve the accuracy of other machine learning algorithms. In this research, we compare the performance of different fuzzy information discrimination measures used as loss functions in the UMAP algorithm while constructing low dimensional embeddings. In order to achieve this, we derive the gradients of the considered losses analytically and employ the Adam algorithm during the loss function optimization process. From the conducted experimental studies we conclude that the use of either the logarithmic fuzzy cross entropy loss without reduced repulsion or the symmetric logarithmic fuzzy cross entropy loss with sufficiently large neighbor count leads to better global structure preservation of the original multidimensional data when compared to the loss function used in the original UMAP algorithm implementation.