Machine Learning (ML) Can Successfully Support Microscopic Differential Counts of Peripheral Blood Smears in a High Throughput Hematology Laboratory

Background: Cytomorphology is the gold standard for quick assessment of peripheral blood and bone marrow samples in hematological neoplasms. It is a broadly-accepted method for orchestrating more specific diagnostics including immunophenotyping or genetics. Inter-/intra-observer-reproducibility of single cell classification is only 75 to 90%. Only a limited number of cells (100 - 500 cells/smear) is read in a time-consuming procedure. Machine learning (ML) is more reliable where human skills are limited, i.e. in handling large amounts of data or images. We here tested ML to differentiate peripheral blood leukocytes in a high throughput hematology laboratory. Aim: To establish an ML-based cell classifier capable of identifying healthy and pathologic cells in digitalized peripheral blood smear scans at an accuracy competitive with or outperforming human expert level. Methods: We selected >2,600 smears out of our unique archive of > 250,000 peripheral blood smears from hematological neoplasms. Depending on quality, we scanned up to 1,000 single cell images per smear. For image acquisition, a Metafer Scanning System (Zeiss Axio Imager.Z2 microscope, automatic slide feeder and automatic oiling device) from MetaSystems (Altlussheim, GER) was used. Areas of interest were defined by pre-scan in 10x magnification followed by high resolution scan in 40x to generate cell images for analysis. Average capture times for 300/500 cells were 3:43/4:37 min We set up a supervised ML-learning model using colour images (144x144 pixels) as input, outputting predicted probabilities of 21 predefined classes. We used ImageNet-pretrained Xception as our base model. We trained, evaluated and deployed the model using Amazon SageMaker on a subset of 82,974 images randomly selected from 514,183 cells captured and labelled for this study. 20 different cell types and one garbage class were classified. We included cell type categories referring to the critical importance of detecting rare leukemia subtypes (e.g. APL). Numbers of images from respective 21 classes ranged from 1,830 to 14,909 (median: 2,945). Minority classes were up-sampledto handle imbalances. Each picture was labelled by highly skilled technicians (median years practicing in this laboratory: 5) and two independent hematologists (median years at microscope: 20). Results: On a separate test set of 8,297 cells, our classifier was able to predict any of the five cell types occurring in the peripheral blood of healthy individuals (PMN, lymphocytes, monocytes, eosinophils, basophils) at very high median accuracy (97.0%) Median prediction accuracy of 15 rare or pathological cell types was 91.3%. For six critical pathological cell forms (myeloblasts, atypical/bilobulated promyelocytes in APL/APLv, hairy cells, lymphoma cells,plasma cells), median accuracy was 93.4% (sensitivity 93.8%). We saw a very high "T98 accuracy" for these cell types (98.5%) which is the accuracy of cell type predictions with prediction probability >0.98 (achieved in 2231/2417 cases), implicating that critical cells predicted with probability <0.98 should be flagged for human expert validation with priority. For all 21 classes median accuracy was 91.7%. Accuracy was lower for cells representing consecutive steps of maturation, e.g. promyelo-/myelo-/metamyelocytes, reproducing inconsistencies from the human-built phenotypic classification system (s.Fig.). Conclusions: We demonstrate an automated workflow using automatic microscopic cell capturing and ML-driven cell differentiation in samples of hematologic patients. Reproducibility, accuracy, sensitivity and specificity are above 90%, for many cell types above 98%. By flagging suspicious cells for humanvalidation, this tool can support even experienced hematology professionals, especially in detecting rare cell types. Given an appropriate scanning speed, it clearly outperforms human investigators in terms of examination time and number of differentiated cells. An ML-based intelligence can make its skills accessible to hematology laboratories on site or after upload of scanned cell images, independent of time/location. A cloud-based infrastructure is available. A prospective head to head challenge between ML-based classifier and human experts comparing sensitivity and accuracy for detection of all cell classes in peripheral blood will be tested to proof suitability for routine use (NCT 4466059). Figure Disclosures Heo: AWS: Current Employment. Wetton:AWS: Current Employment. Drescher:MetaSystems: Current Employment. Hänselmann:MetaSystems: Current Employment. Lörch:MetaSystems: Current equity holder in private company.

Download Full-text

RNA splicing programs define tissue compartments and cell types at single cell resolution

10.1101/2021.05.01.442281 ◽

2021 ◽

Author(s):

Julia Eve Olivieri ◽

Roozbeh Dehghannasiri ◽

Peter Wang ◽

SoRi Jang ◽

Antoine de Morree ◽

...

Keyword(s):

Gene Expression ◽

Single Cell ◽

High Throughput ◽

Rna Splicing ◽

Single Cells ◽

Cell Types ◽

Mouse Lemur ◽

Cell Type ◽

Multiple Organs ◽

Single Cell Pcr

More than 95% of human genes are alternatively spliced. Yet, the extent splicing is regulated at single-cell resolution has remained controversial due to both available data and methods to interpret it. We apply the SpliZ, a new statistical approach that is agnostic to transcript annotation, to detect cell-type-specific regulated splicing in > 110K carefully annotated single cells from 12 human tissues. Using 10x data for discovery, 9.1% of genes with computable SpliZ scores are cell-type specifically spliced. These results are validated with RNA FISH, single cell PCR, and in high throughput with Smart-seq2. Regulated splicing is found in ubiquitously expressed genes such as actin light chain subunit MYL6 and ribosomal protein RPS24, which has an epithelial-specific microexon. 13% of the statistically most variable splice sites in cell-type specifically regulated genes are also most variable in mouse lemur or mouse. SpliZ analysis further reveals 170 genes with regulated splicing during sperm development using, 10 of which are conserved in mouse and mouse lemur. The statistical properties of the SpliZ allow model-based identification of subpopulations within otherwise indistinguishable cells based on gene expression, illustrated by subpopulations of classical monocytes with stereotyped splicing, including an un-annotated exon, in SAT1, a Diamine acetyltransferase. Together, this unsupervised and annotation-free analysis of differential splicing in ultra high throughput droplet-based sequencing of human cells across multiple organs establishes splicing is regulated cell-type-specifically independent of gene expression.

Download Full-text

Reference Transcriptomes of Porcine Peripheral Immune Cells Created Through Bulk and Single-Cell RNA Sequencing

Frontiers in Genetics ◽

10.3389/fgene.2021.689406 ◽

2021 ◽

Vol 12 ◽

Author(s):

Juber Herrera-Uribe ◽

Jayne E. Wiarda ◽

Sathesh K. Sivasankaran ◽

Lance Daharsh ◽

Haibo Liu ◽

...

Keyword(s):

T Cells ◽

Single Cell ◽

Immune Cells ◽

Peripheral Blood ◽

Cell Types ◽

Cell Populations ◽

Rna Seq ◽

Cell Type ◽

Transcriptomic Data ◽

Single Cell Rna Sequencing

Pigs are a valuable human biomedical model and an important protein source supporting global food security. The transcriptomes of peripheral blood immune cells in pigs were defined at the bulk cell-type and single cell levels. First, eight cell types were isolated in bulk from peripheral blood mononuclear cells (PBMCs) by cell sorting, representing Myeloid, NK cells and specific populations of T and B-cells. Transcriptomes for each bulk population of cells were generated by RNA-seq with 10,974 expressed genes detected. Pairwise comparisons between cell types revealed specific expression, while enrichment analysis identified 1,885 to 3,591 significantly enriched genes across all 8 cell types. Gene Ontology analysis for the top 25% of significantly enriched genes (SEG) showed high enrichment of biological processes related to the nature of each cell type. Comparison of gene expression indicated highly significant correlations between pig cells and corresponding human PBMC bulk RNA-seq data available in Haemopedia. Second, higher resolution of distinct cell populations was obtained by single-cell RNA-sequencing (scRNA-seq) of PBMC. Seven PBMC samples were partitioned and sequenced that produced 28,810 single cell transcriptomes distributed across 36 clusters and classified into 13 general cell types including plasmacytoid dendritic cells (DC), conventional DCs, monocytes, B-cell, conventional CD4 and CD8 αβ T-cells, NK cells, and γδ T-cells. Signature gene sets from the human Haemopedia data were assessed for relative enrichment in genes expressed in pig cells and integration of pig scRNA-seq with a public human scRNA-seq dataset provided further validation for similarity between human and pig data. The sorted porcine bulk RNAseq dataset informed classification of scRNA-seq PBMC populations; specifically, an integration of the datasets showed that the pig bulk RNAseq data helped define the CD4CD8 double-positive T-cell populations in the scRNA-seq data. Overall, the data provides deep and well-validated transcriptomic data from sorted PBMC populations and the first single-cell transcriptomic data for porcine PBMCs. This resource will be invaluable for annotation of pig genes controlling immunogenetic traits as part of the porcine Functional Annotation of Animal Genomes (FAANG) project, as well as further study of, and development of new reagents for, porcine immunology.

Download Full-text

Analysis of single cell data as it relates to aging and longevity

Innovation in Aging ◽

10.1093/geroni/igab046.2533 ◽

2021 ◽

Vol 5 (Supplement_1) ◽

pp. 677-678

Author(s):

Tanya Karagiannis ◽

Todd Dowrey ◽

Carlos Villacorta-Martin ◽

George Murphy ◽

Stefano Monti ◽

...

Keyword(s):

Gene Expression ◽

Immune System ◽

Single Cell ◽

Peripheral Blood ◽

Age Groups ◽

Cell Types ◽

Cell Type ◽

Younger Age ◽

Cell Type Specific ◽

Extreme Longevity

Abstract Age-related disability and diseases are known to be delayed in people living to 100 years or more. Changes in the immune system with age are known, including in cell type composition and gene expression differences. To further explore changes in extreme longevity subjects, we investigated peripheral blood immune system cell subpopulations across age and extreme longevity at a single cell resolution. We performed an integrative analysis of public scRNA-seq datasets to define consensus cell types of longevity and age, and classified cell types in our novel New England Centenarian Study dataset. We integrated these datasets together to investigate cell type specific differences at a composition and gene expression level. Our findings identified higher cell type diversity in extreme longevity subjects compared to younger age groups, but no significant difference among younger age groups demonstrating that overall composition differences are unique to longevity. We identified novel differences in myeloid and lymphocyte populations; Extreme longevity subjects have higher composition of CD14+ Monocytes, Natural Killer cells, and T gamma delta populations and lower composition of CD16+ Monocytes and dendritic populations. We characterized gene expression differences between extreme longevity and younger age groups and differences in aging across younger age groups. We found that extreme longevity cell type specific signatures overlapped with the aging signatures by at least 50%. We identified unique genes to extreme longevity that are enriched for pathways specific to immune activation and inflammation, suggesting a protective mechanism for centenarians through efficient activation and regulation of immune subpopulations in peripheral blood.

Download Full-text

Deep learning does not outperform classical machine learning for cell-type annotation

10.1101/653907 ◽

2019 ◽

Cited By ~ 7

Author(s):

Niklas D. Köhler ◽

Maren Büttner ◽

Fabian J. Theis

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Single Cell ◽

Language Processing ◽

Word Identification ◽

Expression Profiles ◽

Gene Expression Profiles ◽

Cell Types ◽

Learning Approaches ◽

Cell Type

AbstractDeep learning has revolutionized image analysis and natural language processing with remarkable accuracies in prediction tasks, such as image labeling or word identification. The origin of this revolution was arguably the deep learning approach by the Hinton lab in 2012, which halved the error rate of existing classifiers in the then 2-year-old ImageNet database1. In hindsight, the combination of algorithmic and hardware advances with the appearance of large and well-labeled datasets has led up to this seminal contribution.The emergence of large amounts of data from single-cell RNA-seq and the recent global effort to chart all cell types in the Human Cell Atlas has attracted an interest in deep-learning applications. However, all current approaches are unsupervised, i.e., learning of latent spaces without using any cell labels, even though supervised learning approaches are often more powerful in feature learning and the most popular approach in the current AI revolution by far.Here, we ask why this is the case. In particular we ask whether supervised deep learning can be used for cell annotation, i.e. to predict cell-type labels from single-cell gene expression profiles. After evaluating 6 classification methods across 14 datasets, we notably find that deep learning does not outperform classical machine-learning methods in the task. Thus, cell-type prediction based on gene-signature derived cell-type labels is potentially too simplistic a task for complex non-linear methods, which demands better labels of functional single-cell readouts. We, therefore, are still waiting for the “ImageNet moment” in single-cell genomics.

Download Full-text

NS-Forest: A machine learning method for the objective identification of minimum marker gene combinations for cell type determination from single cell RNA sequencing

10.1101/2020.09.23.308932 ◽

2020 ◽

Author(s):

Brian Aevermann ◽

Yun Zhang ◽

Mark Novotny ◽

Trygve Bakken ◽

Jeremy Miller ◽

...

Keyword(s):

Machine Learning ◽

Single Cell ◽

Rna Sequencing ◽

Marker Gene ◽

Cell Types ◽

Biological Research ◽

Marker Genes ◽

Cell Type ◽

Type Identity ◽

Wide Range

AbstractSingle cell genomics is rapidly advancing our knowledge of cell phenotypic types and states. Driven by single cell/nucleus RNA sequencing (scRNA-seq) data, comprehensive atlas projects covering a wide range of organisms and tissues are currently underway. As a result, it is critical that the cell transcriptional phenotypes discovered are defined and disseminated in a consistent and concise manner. Molecular biomarkers have historically played an important role in biological research, from defining immune cell-types by surface protein expression to defining diseases by molecular drivers. Here we describe a machine learning-based marker gene selection algorithm, NS-Forest version 2.0, which leverages the non-linear attributes of random forest feature selection and a binary expression scoring approach to discover the minimal marker gene expression combinations that precisely captures the cell type identity represented in the complete scRNA-seq transcriptional profiles. The marker genes selected provide a barcode of the necessary and sufficient characteristics for semantic cell type definition and serve as useful tools for downstream biological investigation. The use of NS-Forest to identify marker genes for human brain middle temporal gyrus cell types reveals the importance of cell signaling and non-coding RNAs in neuronal cell type identity.

Download Full-text

treekoR: identifying cellular-to-phenotype associations by elucidating hierarchical relationships in high-dimensional cytometry data

Genome Biology ◽

10.1186/s13059-021-02526-5 ◽

2021 ◽

Vol 22 (1) ◽

Author(s):

Adam Chan ◽

Wei Jiang ◽

Emily Blyth ◽

Jean Yang ◽

Ellis Patrick

Keyword(s):

Single Cell ◽

Case Studies ◽

High Throughput ◽

Hierarchical Structures ◽

Cell Types ◽

Unsupervised Clustering ◽

High Dimensional ◽

Cell Type ◽

Clustering Techniques ◽

Do So

AbstractHigh-throughput single-cell technologies hold the promise of discovering novel cellular relationships with disease. However, analytical workflows constructed for these technologies to associate cell proportions with disease often employ unsupervised clustering techniques that overlook the valuable hierarchical structures that have been used to define cell types. We present treekoR, a framework that empirically recapitulates these structures, facilitating multiple quantifications and comparisons of cell type proportions. Our results from twelve case studies reinforce the importance of quantifying proportions relative to parent populations in the analyses of cytometry data — as failing to do so can lead to missing important biological insights.

Download Full-text

Identification of cell-type marker genes from plant single-cell RNA-seq data using machine learning

10.1101/2020.11.22.393165 ◽

2020 ◽

Author(s):

Haidong Yan ◽

Qi Song ◽

Jiyoung Lee ◽

John Schiefelbein ◽

Song Li

Keyword(s):

Machine Learning ◽

Random Forest ◽

Single Cell ◽

Correlation Method ◽

Cell Types ◽

Support Vector ◽

Marker Genes ◽

Specific Cell ◽

Sequencing Analysis ◽

Cell Type

AbstractAn essential step of single-cell RNA sequencing analysis is to classify specific cell types with marker genes in order to dissect the biological functions of each individual cell. In this study, we integrated five published scRNA-seq datasets from the Arabidopsis root containing over 25,000 cells and 17 cell clusters. We have compared the performance of seven machine learning methods in classifying these cell types, and determined that the random forest and support vector machine methods performed best. Using feature selection with these two methods and a correlation method, we have identified 600 new marker genes for 10 root cell types, and more than 70% of these machine learning-derived marker genes were not identified before. We found that these new markers not only can assign cell types consistently as the previously known cell markers, but also performed better than existing markers in several evaluation metrics including accuracy and sensitivity. Markers derived by the random forest method, in particular, were expressed in 89-98% of cells in endodermis, trichoblast, and cortex clusters, which is a 29-67% improvement over known markers. Finally, we have found 111 new orthologous marker genes for the trichoblast in five plant species, which expands the number of marker genes by 58-170% in non-Arabidopsis plants. Our results represent a new approach to identify cell-type marker genes from scRNA-seq data and pave the way for cross-species mapping of scRNA-seq data in plants.

Download Full-text

Reference transcriptomes of porcine peripheral immune cells created through bulk and single-cell RNA sequencing

10.1101/2021.04.02.438107 ◽

2021 ◽

Author(s):

Juber Herrera-Uribe ◽

Jayne E Wiarda ◽

Sathesh K Sivasankaran ◽

Lance Daharsh ◽

Haibo Liu ◽

...

Keyword(s):

T Cells ◽

Single Cell ◽

Immune Cells ◽

Peripheral Blood ◽

Cell Types ◽

Cell Populations ◽

Rna Seq ◽

Cell Type ◽

Transcriptomic Data ◽

Single Cell Rna Sequencing

Pigs are a valuable human biomedical model and an important protein source supporting global food security. The transcriptomes of peripheral blood immune cells in pigs were defined at the bulk cell-type and single cell levels. First, eight cell types were isolated in bulk from peripheral blood mononuclear cells (PBMCs) by cell sorting, representing Myeloid, NK cells and specific populations of T and B cells. Transcriptomes for each bulk population of cells were generated by RNA-seq with 10,974 expressed genes detected. Pairwise comparisons between cell types revealed specific expression, while enrichment analysis identified 1,885 to 3,591 significantly enriched genes across all 8 cell types. Gene Ontology analysis for the top 25% of significantly enriched genes (SEG) showed high enrichment of biological processes related to the nature of each cell type. Comparison of gene expression indicated highly significant correlations between pig cells and corresponding human PBMC bulk RNA-seq data available in Haemopedia. Second, higher resolution of distinct cell populations was obtained by single-cell RNA-sequencing (scRNA-seq) of PBMC. Seven PBMC samples were partitioned and sequenced that produced 28,810 single cell transcriptomes distributed across 36 clusters and classified into 13 general cell types including plasmacytoid dendritic cells (DC), conventional DCs, monocytes, B cell, conventional CD4 and CD8 αβ T cells, NK cells, and γδ T cells. Signature gene sets from the human Haemopedia data were assessed for relative enrichment in genes expressed in pig cells and integration of pig scRNA-seq with a public human scRNA-seq dataset provided further validation for similarity between human and pig data. The sorted porcine bulk RNAseq dataset informed classification of scRNA-seq PBMC populations; specifically, an integration of the datasets showed that the pig bulk RNAseq data helped define the CD4CD8 double-positive T cell populations in the scRNA-seq data. Overall, the data provides deep and well-validated transcriptomic data from sorted PBMC populations and the first single-cell transcriptomic data for porcine PBMCs. This resource will be invaluable for annotation of pig genes controlling immunogenetic traits as part of the porcine Functional Annotation of Animal Genomes (FAANG) project, as well as further study of, and development of new reagents for, porcine immunology.

Download Full-text

Identifying cellular-to-phenotype associations by elucidating hierarchical relationships in high-dimensional cytometry data

10.1101/2021.07.08.451609 ◽

2021 ◽

Author(s):

Adam S Chan ◽

Wei Jiang ◽

Emily Blyth ◽

Jean Yee Hwa Yang ◽

Ellis Patrick

Keyword(s):

Single Cell ◽

Case Studies ◽

Hierarchical Structure ◽

High Throughput ◽

Cell Types ◽

Unsupervised Clustering ◽

High Dimensional ◽

Cell Type ◽

Parent Population ◽

Do So

High-throughput single cell technologies hold the promise of discovering novel cellular relationships with disease and necessitate the use of effective analytical workflows. When manual gating is used to define cell types, the gating hierarchy can be used to identify cell types whose abundances change relative to a parent population. This strategy allows subtle changes to be observed that could be missed if small subsets were compared to all measured cells. However, typical analyses that employ unsupervised clustering overlook the valuable hierarchical structure present in cell type definitions by exclusively quantifying the proportions of cell type clusters relative to all cells. We present treekoR, a framework that facilitates multiple quantifications and comparisons of cell type proportions. Our results from twelve case studies reinforce the importance of quantifying proportions relative to parent populations in the analyses of cytometry data - as failing to do so can lead to missing important biological insights.

Download Full-text