Challenges for Computational Stem Cell Biology: A Discussion for the Field

Abstract

The first meetup for Computational Stem Cell Biologists was held at the 2020 annual meeting of the International Society for Stem Cell Research. The discussions highlighted opportunities and barriers to computational stem cell research that require coordinated action across the stem cell sector. The first meetup for Computational Stem Cell Biologists was held at the 2020 annual meeting of the International Society for Stem Cell Research. The discussions highlighted opportunities and barriers to computational stem cell research that require coordinated action across the stem cell sector. The 2020 International Society for Stem Cell Research (ISSCR) annual meeting demonstrated just how integral Computational Stem Cell Biology (CSCB) has become in the stem cell laboratory. This meeting highlighted that it is the researchers who are combining experimental and computational techniques who are driving the evolution of the stem cell field. This was exemplified by Aviv Regev’s presentation in the first plenary session, where she used rich single-cell atlases to computationally infer developmental trajectories and deduce cell lineage maps. Several plenary speakers including Allon Klein, the recipient of the 2020 Dr. Susan Lim Outstanding Young Investigator Award, showcased a variety of computational approaches to better understand stem cell biology. The 2020 ISSCR annual meeting also hosted a virtual networking event for researchers interested in CSCB. It attracted >130 attendees interested in developments across the discipline, with 44 of those actively engaging in the discussion. The session was meant to serve as a forum to connect researchers across geographical and disciplinary boundaries, and also to serve as an informal survey to identify the most prominent short-term and long-term questions, issues, and challenges that the field faces (as described in detail below). We were glad to see a large spread in the career stage and expertise of attendees, stretching from PIs “developing experimental and computational techniques to better understand cell reprogramming” to post-docs who were “stem cell biologists who sometimes wish to be computational biologists” and Ph.D. students who were “starting in September and would love to learn about computational biology”. As CSCB participant numbers increase, so has the need to establish the community more formally. Here we report a summary of the topics and recommendations for development of CSCB within the ISSCR community. The session tackled four main questions designed to stretch the discussion from an understanding of where we are now as a field to where we think we should go next and how we can ensure that we have the right network to achieve this. When considering the biggest barriers to CSCB research, the question of how to find, reuse, or combine the vast amount of data already produced by stem cell researchers was a recurring theme. A computational researcher wishing to study a particular stem cell question must search through numerous databases, which lack sufficient information, to systematically (1) identify samples from the cell types of interest; (2) refine the search by properties such as disease status, sex, age/developmental stage, or genetic variants; (3) filter by experimental treatments and conditions; and (4) record genetic modification (e.g., reporter gene constructs or gene editing). Even once the studies have been identified, locating all of the data is difficult, as these may be in multiple resources, or equally problematically, replicated in multiple resources without adequate mapping between them. Reanalysis or meta-analysis of combined studies can yield new insights into a system, but curating data for this purpose remains a difficult task: as one participant commented, “there has already been a disproportionate expenditure of resources available to generate data describing stem cell models, without also investing in ways to ensure that we can use these data effectively”—and many in the discussion felt that this is a mission for the wider stem cell society. Finding the right reference data to classify cell types and differentiation stages obtained in stem cell cultures is particularly fraught in the absence of high-quality developmental cell atlases. This was most obvious to the group for key tool development areas such as cell identity/classification, cell fate prediction, and cellular engineering. However, cellular types and states cannot be assigned using a reference if they have not been well characterized previously. “Ground truth” datasets are needed for evaluation of computational tools seeking to model pluripotent networks or predict cell fate change. Dynamical data that are the most useful for prediction are particularly rare. To date, large-scale atlas initiatives, such as ENCODE (Moore et al., 2020Moore J.E. Purcaro M.J. Pratt H.E. Epstein C.B. Shoresh N. Adrian J. Kawli T. Davis C.A. Dobin A. Kaul R. et al.ENCODE Project ConsortiumExpanded encyclopaedias of DNA elements in the human and mouse genomes.Nature. 2020; 583: 699-710Crossref PubMed Scopus (355) Google Scholar) and FANTOM (Forrest et al., 2014Forrest A.R.R. Kawaji H. Rehli M. Baillie J.K. de Hoon M.J.L. Haberle V. Lassmann T. Kulakovskiy I.V. Lizio M. Itoh M. et al.FANTOM Consortium and the RIKEN PMI and CLST (DGT)A promoter-level mammalian expression atlas.Nature. 2014; 507: 462-470Crossref PubMed Scopus (1179) Google Scholar), have primarily sampled mature tissue types. We note that the Human Cell Atlas has begun to include developmental stages for some tissues (cf. Bock et al., 2020Bock C. Boutros M. Camp J.G. Clarke L. Clevers H. Knoblich J.A. Liberali P. Regev A. Rios A.C. Stegle O. et al.The Organoid Cell Atlas: A Rosetta Stone for Biomedical Discovery and Regenerative Therapy.Zenodo. 2020; https://doi.org/10.5281/zenodo.4001717Crossref Google Scholar), as well as a recent developmental atlas (Cao et al., 2020Cao J. O’Day D.R. Pliner H.A. Kingsley P.D. Deng M. Daza R.M. Zager M.A. Aldinger K.A. Blecher-Gonen R. Zhang F. et al.A human cell atlas of fetal gene expression.Science. 2020; 370: eaba7721Crossref PubMed Scopus (117) Google Scholar; Domcke et al., 2020Domcke S. Hill A.J. Daza R.M. Cao J. O’Day D.R. Pliner H.A. Aldinger K.A. Pokholok D. Zhang F. Milbank J.H. et al.A human cell atlas of fetal chromatin accessibility.Science. 2020; 370: eaba7612Crossref PubMed Scopus (83) Google Scholar). There was strong endorsement from participants for a stem cell atlas project to create suitable reference data from differentiating stem cell lines for comparison across multiple -omic technologies. In some instances the appropriate reference data simply do not exist; these gaps should be recognized and funded accordingly. The CSCB community wanted to see atlas efforts make use of global stem cell collections that are genotyped and phenotyped, such as the HipSci (Streeter et al., 2017Streeter I. Harrison P.W. Faulconbridge A. Flicek P. Parkinson H. Clarke L. The HipSci ConsortiumThe human-induced pluripotent stem cell initiative-data resources for cellular genetics.Nucleic Acids Res. 2017; 45: D691-D697Crossref PubMed Scopus (51) Google Scholar) or CiRA Foundation iPS Cell Stock (Umekage et al., 2019Umekage M. Sato Y. Takasu N. Overview: an iPS cell stock at CiRA.Inflamm. Regen. 2019; 39: 17Crossref PubMed Scopus (49) Google Scholar). This would serve the dual purpose of building a rich genotype-phenotype catalog associated with publicly available lines. FAIR data is a principle adopted by major funders and by major consortia (Box 1). But what does FAIR mean for the stem cell field? Reusing data that have been created by individual researchers was identified as both an opportunity and a major challenge for the field. It is hard to find relevant and high-quality examples of specific stem cell or developmental stages. Platforms such as Stemformatics, which focus on data curation of public stem cell transcriptome experiments, apply QC metrics that fail 30% of data reviewed from the public domain. Several such niche resources for relevant data exist (e.g. Stemformatics [Choi et al., 2019Choi J. Pacheco C.M. Mosbergen R. Korn O. Chen T. Nagpal I. Englart S. Angel P.W. Wells C.A. Stemformatics: visualize and download curated stem cell data.Nucleic Acids Res. 2019; 47: D841-D846Crossref PubMed Scopus (14) Google Scholar] and the human pluripotent stem cell registry [Mah et al., 2020Mah N. Seltmann S. Aran B. Steeg R. Dewender J. Bultjer N. Veiga A. Stacey G.N. Kurtz A. Access to stem cell data and registration of pluripotent cell lines: The Human Pluripotent Stem Cell Registry (hPSCreg).Stem Cell Res. (Amst.). 2020; 47: 101887Crossref PubMed Scopus (8) Google Scholar]), but what is needed is interoperability between these resources to aid in data sharing and uptake by the community. Metadata is not interoperable in the stem cell field, and we fail to systematically capture the most relevant information in standardized formats, including naming conventions for stem cell lines. Widespread and coordinated adoption of FAIR data practices could accomplish a level of semantic interoperability that would not only enable researchers to find the right kind of data, but to query catalogs of metadata in a machine-readable way. It would then be possible to find suitable data for compilation into well-characterized reference datasets that could serve as benchmarks for the community.Box 1Fair Data PrinciplesFindable: descriptions of how the data were generated and processed are complete, and standardized ontologies or other annotations are used to describe the experimental system.Accessible: data and metadata are curated in public repositories.Interoperable: data and metadata are in community-agreed-upon formats.Reusable: data, metadata, and code are provided. (Wilkinson et al., 2016Wilkinson M.D. Dumontier M. Aalbersberg I.J.J. Appleton G. Axton M. Baak A. Blomberg N. Boiten J.-W. da Silva Santos L.B. Bourne P.E. et al.The FAIR Guiding Principles for scientific data management and stewardship.Sci. Data. 2016; 3: 160018Crossref PubMed Scopus (4424) Google Scholar.) Findable: descriptions of how the data were generated and processed are complete, and standardized ontologies or other annotations are used to describe the experimental system. Accessible: data and metadata are curated in public repositories. Interoperable: data and metadata are in community-agreed-upon formats. Reusable: data, metadata, and code are provided. (Wilkinson et al., 2016Wilkinson M.D. Dumontier M. Aalbersberg I.J.J. Appleton G. Axton M. Baak A. Blomberg N. Boiten J.-W. da Silva Santos L.B. Bourne P.E. et al.The FAIR Guiding Principles for scientific data management and stewardship.Sci. Data. 2016; 3: 160018Crossref PubMed Scopus (4424) Google Scholar.) A second major barrier was “how to define a stem cell or derived cell type.” The implicit assumption that cell types are static entities is largely a byproduct of how we observe cellular systems rather than a ground truth of the systems themselves. Traditional notions of cell types defined by morphology and histology need to be reconciled with the signatures derived from computational analysis of -omic data. Computational approaches to labeling groups of cells include (1) annotation with prior knowledge—such as the presence of a validated lineage marker, (2) comparison to reference sets drawn from the reference databases discussed above, and (3) machine-learning-based annotation. Marker-based methods are the most conventional way to define a cell type, and since these markers also can be used for cell sorting and isolation, the marker-based cluster labeling supports integrity with subsequent experiments. However, the employment of canonical markers relies on background knowledge of scientists and is highly variable across laboratories. Furthermore, it has become obvious that cellular phenotypes are more heterogeneous than cell types previously defined by the marker expression. As a result, standard cell ontologies that rely on capture or and lineage methods are of developmental is needed to a of cell into cell morphology and and but these should also for the and of new cell types and cell the of is the of differentiation are within cell lines and between is a so the is to computationally these from the data the between to and is not and for would computational of cell types and cell states et al., C.A. can in research PubMed Scopus Google Scholar). The field would be by to highly characterized stem cell lines that can be used for more of CSCB data. It was by the cell in the discussion that of cell lines is In cell need for and of metadata, including cell with sufficient detail to enable comparison of data across laboratories. a are studies where such as have been systematically and the of the stem cell The lack of data on well-characterized cell lines the field with information about how to on new lines or The of and on cell lines that is by the for Cell is an of We that the ISSCR could a to on the to describe a stem cell The discussion on The first was the of approaches to understanding and developmental The second was better understanding of the to models, particularly combining data from or the development of cell tools is a highly research the field is to needed to or labeling of cell such as using the analysis for cluster prediction and cluster is a are for to find cells of a and to between cells of but the of the the are is This to an in the by many by annotation of or QC and in of the cell it is hard to apply generated by to new so the derived a computationally derived truth” of cell remains The discussion highlighted the need for computational tools to information from to understand Computational approaches for cell lineage are in and with data some recent developments of the et al., C. D. J. P. S. M. The and of cell fate are by of 2014; PubMed Scopus Google Scholar), they cells into a if is to enable of if those cells should be using a tools rely on data which may not have sufficient information to deduce the between Furthermore, mapping the in networks or better of and or chromatin as well as of in a The computational challenge is from these combined data. in cell as a a of cell states within cell type, a cell may require a that the and and the of a The of datasets is an that new to and for it difficult to data from and the in that be from the In some it is to generate more data than this the would be to on new and data types and data atlases such as the ConsortiumThe human at cellular the Human Atlas 2019; PubMed Scopus Google Scholar) data and computationally data a reference (e.g. et al., T. A. P. C. Y. M. P. R. of 2019; PubMed Scopus Google Scholar). is which in in data It is possible to the in a to such as and or for this. However, these methods should be used as they could relevant data, or if may where exist Wells and C.A. J. of Stem from to Cell 2019; PubMed Scopus (4) Google for with the of single-cell is a in tools (e.g., et al., T. A. P. C. Y. M. P. R. of 2019; PubMed Scopus Google Scholar] and et al., B. B. B. of heterogeneous single-cell using 2019; PubMed Scopus Google that the of data in of the of approaches are useful to find in data but do so by some as This on to understand the between and cellular it is difficult to from (e.g., in the of should be in these data, and computational tools should to some of these As stem cell we have as one of a to better understand and and stem cells It was to see the approaches to understanding cell phenotypes the to lineage and chromatin and gene in with the on data it be to how we combine these to a of cellular and the of new cell of the of CSCB is the that the tools and should understanding of stem cell systems and where possible as we and experiments, must be into to be available for human The question of how to the research experimental data and in and must be if the stem cell community is to knowledge from of these Even as the data that we more we are to that for of data on large numbers of The was that we have a as a community to the most from these datasets and use to go into the than has been previously in which we to see is in the and analysis of technologies. that include for which are in cell be a vast the of from gene expression data. A aid development of more differentiation from a It also questions on cell cell differentiation and The of these datasets new challenges for data particularly the of cell approaches with to cell the and in data. As a we were by the opportunity to computational expertise in cell with expertise in single-cell technologies. opportunities for stem cell and CSCB researchers from stem cell collections that the expression and for human stem cells and CSCB researchers must be in to ensure that the experimental for data is with the data There is also a opportunity to methods in CSCB. to or cells to stem cell biologists through many or and other to cell Even with a of by prior these experimental approaches require of and resources with Computational approaches that have already in this for et al., P. H. da network to stem cell 2014; PubMed Scopus Google Scholar), et al., J. H. H. C.M. et al.FANTOM computational for between human cell 2016; PubMed Scopus Google Scholar), and et al., A. S. G. A. A for in Cell PubMed Scopus Google Scholar). single-cell such as can identify of cell associated with specific new to cells a specific these approaches not cell yield if the cell is not within the cells networks and may in new cell types or cell it is possible to find such derived it be in understanding of how networks cellular could an opportunity to single-cell differentiation and methods have been used to predict single-cell et al., M. single-cell 2019; PubMed Scopus Google Scholar). The of these approaches to predict new to cell remains to be A for such an was to into for development of the CSCB community is it is that we can identify how to the field and make this for new researchers the field. where was was that CSCB needed to across the and of this A understanding of not only for the development of the most useful tools but also of or As of this it is also the of the CSCB community to better other of the stem cell community about the and as well as the of the of tools that are The of available tools to et al., L. B. A. the single-cell analysis with the PubMed Scopus Google Scholar) is an to this. new this of tools can be can the new tools than these It may also be for the CSCB community to a “starting of tools for in stem cell biology. of the tools that are by the CSCB community can be to but to do so with a understanding of the data can to biologists the CSCB field, it is to as and understand an is used and how are We that the of CSCB research to the needed to in the of We were by the of researchers to understanding of computational this we building opportunities for the CSCB community to for the knowledge and also the of as it that may in data. this in some of the and that CSCB for CSCB in this field should to be in a way and as a need to a of and and CSCB PIs need to and in at one and developmental and stem cell with biologists from all and data with from other (e.g. systems or and project management of opportunities for scientists to learn techniques and create an where knowledge sharing is and find for computational in this field should to be in a way and as a need to a of and and CSCB PIs need to this. and in at one and developmental and stem cell biology. with biologists from all and data with from other (e.g. systems or and project management of opportunities for scientists to learn techniques and create an where knowledge sharing is and find for computational It was also felt that the computational of the community was a need to better understand the of A with on the of is that the of with the need for it was felt that many single-cell studies are not of the of the in the of the tools used for It be as a community to not only tools but also a of practices to those used in other areas of the stem cell community to ensure that analysis a of This that the can be more and be more to be many now require more descriptions of the methods As a we should to computational resources, by and This with a of computational and In a strong community also be to the development of new The meeting of computational stem cell biologists was a in researchers from and with a variety of for computational stem cell The discussions the to the computational community within the ISSCR and There was for a CSCB in the to researchers from both the and to about in and using CSCB There are also to some and on topics to both the stem cell community as a and also ensure that the practices are used In summary the from this meeting can be as is a and need to better for data and that is the of the stem cell community as a to The CSCB community be the ISSCR to that this and on the of is a to include stem cells more the cell atlas where see these with stem cell resources where of the lines has already was felt that the of and the development of computational tools should within the stem cell community. with this it was felt that a to both and use tools in a way that and should become more a for the community. In the way that as a stem cell it is to identify of stem the CSCB community should be to identify and the or of computational a methods development of a of was in with datasets and a static of cell As the amount of data the to create tools also new areas of stem cell and the need for and data We would to the ISSCR for this meetup and all the participants for It is an for the CSCB and we to is a scientific and of a cell is an with Stem Cell is an of the these and do not the or of or the other have a of to

Topics & Concepts