Curating OmixAtlas - Single Cell

Single Cell RNASeq Data

1. Dataset Level Metadata

Field	Description	Ontology	GUI Display Name	Polly-Python Display Name
Organism	This field represents the organism from which the samples originated. Organism labels already present in the source metadata are normalized using a normalization model. In case the organism labels are missing, related texts and abstracts are processed and further normalized to get the organism metadata label.	NCBI Taxonomy	Organism	curated_organism
Tissue	This field represents the tissue(s) from which the samples in the dataset are derived. Tissue labels already present in the source metadata are normalized using a normalization model. In cases where tissue labels are missing, related texts and abstracts are processed and further normalized to get the tissue metadata label. Tissue labels for datasets consists of all tissue names from which the samples are derived. Tissue labels are annotated for samples extracted from a healthy tissue or a diseased tissue. Key specifications for tissue metadata annotations are as follows: All labels are harmonized with Brenda Tissue Ontology Dual Channel datasets where the author has studied two different cases in a single sample are not curated with tissue metadata Datasets with numerical metadata are not annotated	Brenda Tissue Ontology	Tissue	curated_tissue
Drug	This field represents the drug(s) that have been used in the treatment of the samples or relate to the experiment in some other way. Drug labels already present in the source metadata are normalized using a normalization model. In cases where drug labels are missing, related texts and abstracts are processed and further normalized to get the drug metadata label. Drug labels are annotated for the following types of sample treatments: Chemical treatment - Treatment for activation/regulation/inhibition of a protein or nucleic acid. Stimulation - Treatment to elicit an immune response. Eg. Cytokine stimulation (Interferon alpha, TNF-a), LPS Drug treatment - Treatment with a substance used to treat an illness, relieve a symptom or modify a chemical process in the body for a specific purpose. Eg. 3,5-diethoxycarbonyl-1,4-dihydrocolidine (DDC) Drug labels are not annotated for the following types of sample treatments: Control/ Vehicle - The control group receives either no treatment, a standard treatment whose effect is already known, or a placebo. Mostly the control sample contains organic solvents Treatment for genetic perturbation - Treatments used for clonal selection, inducible genetic perturbation. Eg. Puromycin, Tamoxifen, Doxycycline Transfection - RNA molecules (miRNA, shRNA, siRNA etc) for genetic manipulation Other treatments - Treatments such as chemotherapy, radiation therapy or with antibodies Other - Culture media, supplements and detergents. Eg. DMEM, LB Media, Agar Media, Glucose, amino acids, fats, lipids, SDS Note : Any mention of the drug in the text is included as a drug label irrespective of whether it is being used in the experiment or not.	PubChem	Drug	curated_drug
Disease	This field represents the disease(s) being studied in the experiment. Disease labels already present in the source metadata are normalized using a normalization model. In case the disease labels are missing, related texts and abstracts are processed and further normalized to get the disease metadata label. Disease labels are annotated when the samples have been collected from diseased tissue or organism. Examples of such cases are as follows: If the cell line has been extracted from a diseased organism/tissue and been immortalised for further study then it will be annotated for the disease. For example, if the samples are extracted from an osteosarcoma tumor then disease annotation for this dataset will be "osteosarcoma" If a disease model has been created through genetic modification or has been artificially created in lab for further studies, then it will be annotated for that disease. For example, a genetically engineered mouse model where an immunodeficient or humanized mouse is implanted with a tissue from a patient's tumor. Note : In studies, where the cell lines are extracted from a healthy tissue/organism and then conditioned to induce disease, the disease label for such a dataset will be "normal" since the sample is not extracted from any diseased tissue or organism. Key specifications for disease metadata annotations are: All types of disease including viral, and bacterial infections, metabolic syndromes such as obesity, and diabetes, cancers such as lung neoplasm, breast neoplasm etc. are included as disease metadata labels. Labels for disease mentioned for both full form and short form are annotated. For example, Acute Myeloid Leukaemia will be labelled for both AML, Myeloid Leukaemia, Leukaemia and Acute Myeloid Leukaemia For a dataset, diseases for each sample in the dataset are annotated Disease mentioned in the metadata of a dataset, irrespective of the study is labelled Processes such as "carcinogenesis" or "tumorigenesis" are not annotated as diseases Development abnormalities are not included Dual Channel datasets where the author has studied two different cases in a single sample are not curated for disease metadata	MeSH	Disease	curated_disease
Cell type	This field represents the cell type of the samples within the study. Cell type labels are annotated in cases where the authors have cultured a particular cell type either extracted from tissues or developmental organs or generated in the lab and then used it in further experiment. Key specifications for cell type metadata annotations are as follows: All labels are harmonized with Cell Ontology Cell lines are not curated for cell types For cell types with functional terms such as circulating, associated, derived, etc., only the cell type is annotated For tissue-specific cell types, the tissue name along with the cell type is labelled. Eg - aortic endothelial cells, spinal motor neurons Organism terms are not included in the cell type labels. Eg. mouse "HSPCs" Abbreviated cell types with functional conditions are labelled as abbreviated terms. Eg. CTCs	Cell Ontology	Cell Type	curated_cell_type
Cell line	This field represents the cell line from which the samples were extracted. List of the population of modified cells used for the study. Cell line labels already present in the source metadata are normalized using synonyms present in the cell line ontology we use. Cell line labels are annotated in cases where the authors have cultured a particular cell line or bought it from organizations such as ATCC and then used in further experiments. Eg. MDAMB-231, HEK-293 Key specifications for cell line metadata annotations are as follows: All labels are harmonized with The Cellosaurus Ontology Dual Channel datasets where the author has studied two different cases in a single sample are not curated for cell line metadata Datasets with numerical metadata are not annotated	The Cellosaurus	Cell Line	curated_cell_line
Other metadata fields
Field	Description	GUI Display Name	Polly-Python Display Name
Abstract	This field provides the abstract of the publication associated with the dataset.	NA	abstract
Year	This field provides the year in which the dataset or study is published.	Year	year
Gene	This field provides the gene(s) studied in the dataset.	Gene	curated_gene
Single cell chemistry	This field represents the sequencing method/platform used for sequencing the single cell genome.	Single cell chemsitry	curated_single_cell_chemistry
Sampling technique	This field represents the method/procedure used for collecting samples.	Sampling technique	curated_sampling_technique
Sampling storage technique	This field represents the method/technique used for preservation or storage of sample for analysis.	Sampling storage technique	curated_storage_technique
Summary	This field provides a detailed summary of the publication (can be the abstract) or a summary of the experiment.	Summary (Available for datasets from GEO only)	summary
Overall design	This field provides information on the overall design of the experiment as given by the author.	Overall Design (Available for datasets from GEO only)	overall_design
Publication	This field provides the link to the publication associated with the dataset. If the associated publication information is not available, then this field provides the link to the data source providing more information on the dataset.	NA	publication
Source	This field provides the name of the source repository from where the dataset is fetched.	Source	dataset_source
Description	This field provides a brief description of the experiment or the study.	Title (view details page & card view) Description (Table view)	description
Data Type	This field provides the type of biomolecular data represented/studied in the dataset.	NA	data_type
Dataset ID	This field provides the unique id for the dataset/study to represent a group of samples.	Dataset ID	dataset_id
Number of Cells	This field represents the number of cells/observations in the dataset.	Number of Cells	total_num_cells
Samples	This field represents the total number of samples in a dataset.	Samples	total_num_samples

2. Sample Level Metadata

Field	Description	Ontology	GUI Display Name	Polly-Python Display Name
Ontology-driven Fields
Tissue	This field represents the tissue(s) from which the samples originated. Tissue labels already present in the source metadata are normalized using a normalization model. In cases where tissue labels are missing, related texts and abstracts are processed and further normalized to get the tissue metadata label. Tissue labels are annotated for samples extracted from healthy or diseased tissue. All labels are harmonized with Brenda Tissue Ontology.	Brenda Tissue Ontology	Tissue	curated_tissue
Disease	At the sample level, this field represents the disease associated with a particular sample. Disease labels already present in the source metadata are normalized using a normalization model. In case the disease labels are missing, related texts and abstracts are processed and further normalized to get the disease metadata label. Disease labels are annotated for a sample when the samples have been collected from diseased tissue or organism. Examples of such cases are as follows: If the cell line has been extracted from a diseased organism/tissue and been immortalised for further study then it will be annotated for the disease. For example, if the samples under study are extracted from osteosarcoma tumors then disease annotation for this sample will be "osteosarcoma" If a disease model has been created through genetic modification or has been artificially created in a lab for further studies, then it will be annotated for that disease. For example, a genetically engineered mouse model where an immunodeficient or humanized mouse is implanted with tissue from a patient's tumor. At the sample level, disease labels are annotated for the following sample type: Clinical- Samples extracted from diseased patients, tissue, cell lines etc. GEM (Genetically Engineered Models) - Samples extracted from genetically engineered mouse models Diet Induced - Samples extracted from diet induced mouse models Xenograft - Samples extracted from patient-derived xenograft or cell line-derived xenograft mouse models Infection - Samples extracted from infected organisms, tissues, cell lines or cultures. Example: Viral infection, Bacterial infection etc Other - Any other type of sample in which disease has not been mentioned in metadata, but is not normal	MeSH	Disease	curated_disease
Drug	This field represents the drugs that have been used in the treatment of a sample. Drug labels already present in the source metadata are normalized using a normalization model. In cases where drug labels are missing, related texts and abstracts are processed and and further normalized to get the drug metadata label. Drug labels are annotated for the following types of sample treatments: Chemical treatment - Treatment for activation/regulation/inhibition of a protein or nucleic acid. Stimulation - Treatment to elicit an immune response. Eg. Cytokine stimulation (Interferon alpha, TNF-a), LPS Drug treatment - Treatment with a substance used to treat an illness, relieve a symptom or modify a chemical process in the body for a specific purpose. Eg. 3,5-diethoxycarbonyl-1,4-dihydrocolidine (DDC) Drug labels are not annotated for the following types of sample treatments: Control/ Vehicle - The control group receives either no treatment, a standard treatment whose effect is already known, or a placebo. Mostly the control sample contains organic solvents Treatment for genetic perturbation - Treatments used for clonal selection, inducible genetic perturbation. Eg. Puromycin, Tamoxifen, Doxycycline Transfection - RNA molecules (miRNA, shRNA, siRNA etc) for genetic manipulation Other treatments - Treatments such as chemotherapy, radiation therapy or with antibodies Other - Culture media, supplements and detergents. Eg. DMEM, LB Media, Agar Media, Glucose, amino acids, fats, lipids, SDS Note: Any mention of the drug in the text is included as a drug label irrespective of whether it is being used in the experiment or not.	PubChem	Drug	curated_drug
Cell line	This field represents the cell line from which the sample was derived. Cell line labels already present in the source metadata are normalized using synonyms present in the cell line ontology we use. The cell line field is curated for a sample if the authors have cultured a particular cell line or bought it from organisations such as ATCC and then used in the further experiment. The names of the cell lines are harmonized by the cellosaurus ontology.	The Cellosaurus	Cell Line	curated_cell_line
Cell Type	This field represents the cell type of the sample. Cell type labels are annotated where the authors have cultured a particular cell type either extracted from tissues or developmental organs or generated in the lab and then used it in the further experiment.The cell type field provides the closest cell type name as per the Cell Ontology. This cell type label can be either source derived or by manual cell type annotation.	Cell Ontology	Cell Type	curated_cell_type
Other Metadata Fields
Field	Description	GUI Display Name	Polly-Python Display Name
Gene	Gene of interest in the sample	Gene	curated_gene
Genetic Modification	This fields represents the kind of genetic modification done on the sample.	Genetic Modification	curated_genetic_modification_type
Modified Genes	This fields represents the gene(s) modified in the sample under study.	NA	curated_gene_modified
Donor Type	This field represents the type/clinical condition of the donor	Donor Type	curated_donor_type
Donor Sample Type	This field represents the location/area from where the tumor samples are collected	Donor Sample Type	curated_donor_sample_type
Gender	This fields represents the gender of the organism from which the sample was derived	Gender	curated_gender
Minimum age	This fields provides the lower limit of the age range of the organism from which the samples have been obtained for the study.	Minimum age	curated_min_age
Maximum age	This fields provides the upper limit of the age range of the organism from which the samples have been obtained for the study.	Maximum age	curated_max_age
Age unit	This fields provides the age unit of the organism from which samples have been obtained. It is years for samples from humans and weeks for samples from mice.	Age unit	curated_age_unit
Sampling site	This fields provides the location/area from where the tumor samples are collected.	Sampling site	curated_sampling_site
Treatment	Name of the treatment given to the samples i.e. name of the chemical/drug/therapy	Treatment Name	curated_treatment_name
Treatment type	The type of treatment given to the sample	Treatment Type	curated_treatment_type
Response to Treatment	This field indicates the type/extent of response on the treatment	Treatment Response	curated_treatment_response
Author Cell Type	This field represents the author cell type as mentioned in the publication associated with the dataset.	Author cell type	curated_raw_cell_type
Marker present	This field represents the gene name/names that are differentially expressed in a cluster based on which the cell type of the cluster is annotated.	Marker Present	curated_marker_present
Marker absent	This field represents the gene name/names that are absent in a cluster based on which the cell type of the cluster is annotated.	Marker Absent	curated_marker_absent
Cell ontology ID	This field represents the unique ID for the cell type according to the Cell Ontology.	NA	curated_cell_ontology_id
Clusters	This field provides the cluster number to which each cell belongs after subjecting the dataset to the clustering process (Using Leiden or other algorithms).	Cell Type Cluster	clusters
UMI Counts	This field represents the Unique Molecular Identifier count per cell. It represents an absolute number of observed transcripts. The number should be higher than 500 in a cell.	UMI Counts	umi_counts
Sample ID	This field represents the unique ID of the sample.	Sample ID	sample_id
Cell ID	This field represents the unique ID associated with every cell	NA	cell_id
Gene Counts	This field represents the number of genes detected per cell (defined as genes with cpm > 1); cpm=counts per million	Gene Counts	gene_counts
Mitochondrial count	This field represents the percentage of mitochondrial counts in total counts for a cell.	NA	percent_mito
Title	This field represents the title of the sample, representing the type/genotype/origin/experimental condition of the sample	NA	title

3. Feature Level Metadata

Field	Description	Polly-Python Display Name
Feature ID	This field represents the ID of the feature (gene, metabolite, protein etc) being measured.	feature_id
Highly variable gene	This field indicates whether the gene is highly variable. For highly variable genes - True; otherwise -False.	highly_variable
Number of cells	This field represents the number of cells which are containing the gene.	n_cells

Cell Type Annotation for single cell datasets: Manual Curation Process

Cell-type labels are assigned at the cell cluster level based on expression signatures using ontology or controlled vocabularies. For datasets, where cell-type annotations are not available from the source (mainly GEO datasets), we manually curate the cell-type information based on the differential marker expression for clusters. In cases where cell type annotations are already available in datasets at source, datasets are not manually re-curated.

Process Flow

Figure 1. Process Flow

I) Manual identification of the cell types and markers from Publications -

Internal curators determine if a particular dataset can be curated for cell type by going through the publications associated with the dataset. In publications, the information on cell type and the corresponding marker is present either in the figures (UMAP, T-SNE plots), text or supplementary files. If this information is not present, then such datasets are marked as 'Not Curatable'. The following types of studies fall under the category of 'Not curatable' datasets.

Single Cell Type Study - The whole study was done on only one cell type.
Lineages - The study included the lineage of one cell type. Example - T helper cells, T memory cells etc.
Cell Cycle Studies - In this study the differential markers were studied for the G1, S, G2 and M phases of the cell cycle for a particular cell type.
Methods - Different analysis methods were studied
Publications having \< 2000 cells - These publications used methods which could not be reproduced. Therefore, these were not curated.
Publication not available
Marker information not available in the publication
Marker Cell Type Info Absent
Marker Info Absent
Time Point
Cell lines
Transitional Cells
Cell Type Info Absent
Embryonic Development
Organoids
Others

II) Metadata addition- Annotation of clusters

The process of cell-type cluster annotation for curatable datasets is based on the general scRNASeq generalworkflow using the Scanpy library with steps as shown below in the figure:

Homepage

Figure 2. Workflow

UMAP/tSNE plots are generated as a result of single-cell raw count processing. By visualization of clusters with UMAP/t-SNE plots, cell type cluster annotation is done.

1. Cluster annotation with raw cell type (cell type terminology used in publication):

Based on the marker expression value for each cluster, the cell type is annotated to the cluster. This annotation is added as a field named curated_raw_cell_type. The raw cell type cluster annotation is compared with cell type annotation from the publication such as:

All or most cell types are annotated
UMAP is structurally similar to the one in the publication
Relative proportions of cell types are matching
Relative positions of cell types are matching
Ontological terms and marker information are added for the cell type

2. Ontological terms and marker information:

Cell type ontology + ontology ID: Cell-type annotation corresponding to Cell Ontology. This is given as the field named: curated_cell_type
Marker information: Gene name/names that are differentially expressed in the cluster
curated_marker_present: Gene name/names that are differentially expressed in the cluster
curated_marker_absent: Gene name/names that are absent in the cluster

NOTE : For manual cell type curation, datasets should be available on Polly.