Scientists want cellxgene to support data modalities that help define cell types and states

[ambrosejcarr]

Background

The central dogma of molecular biology describes the flow of molecular information between genetic systems. Each of these stages and processes can be assayed to generate count data. In brief:

Input    Output               Process                            Multiplicity
DNA      poised polymerase    transcription initiation           1:N
pp.      preRNA               transcription                      1:1
preRNA   mRNA                 splicing,                          1:N
mRNA     protein              translation,                       1:1           
protein  active protein       phosphorylation, ubiquitinlyation, 1:N           
                              cleavage, other, )

There are common assays that generate biological data on molecules (not exhaustive):

Molecule  Assays
DNA       ATAC-seq, DNAse-seq
pp.       Polymerase ChIP
preRNA    end-localized RNAseq: 10x 3', 10x 5', Drop-seq, InDrops, Full length RNAseq: SS2
mRNA      end-localized RNAseq: 10x 3', 10x 5', Drop-seq, InDrops, Full length RNAseq: SS2
protein   CITE-seq, MIBI, CyTOF, 

There are also assays or algorithms that measure or infer processes (not exhaustive):

Process      Assay
splicing     RNA-velocity
translation  ribosome profiling

Important notes:

• Later stages of genetic information transfer can be translated, usually with a loss of information, to earlier states. For example, there is a 1:1 correspondence between active protein > mRNA > preRNA > DNA. The converse is not true.
• "molecules per gene", the most common measurement from 10x genomics, records information that corresponds to "pre-RNA".
• Most RNA-seq assays can provide information on transcripts. They elect not to because the data have low accuracy; they don't record enough of the gene to identify all the splicing events and therefore often cannot uniquely identify an mRNA molecule from the set of possible molecules.

The types of assays that users have requested are listed as child tickets of this epic.

[ambrosejcarr]

chanzuckerberg/cellxgene#1673 contains a potential implementation for multiple symbol types, which could be relevant to datasets with multiple modalities that map to the same feature types. e.g. PTPRC (gene) vs CD45R0 (messenger, protein)