Scientist Stories: John Marioni, Gene Expression in Single-Cells

So, in this brief presentation, what I would like to do, though, is talk about something a little bit different, and that is to give, on behalf of myself, Anthony Philippakis, Ben Petten, and Jeremy Freeman, an update on the Data Coordination Platform. So, what, you might ask, is the Data Coordination Platform, and why do we need it? It should be self-evident that in a project of this scale, we're going to generate simply vast amounts of data. And if we're going to make sense of that vast amount of data, if we're going to use it properly, that data needs to be accessible. So what we want to do simply via this platform is to provide all researchers throughout the world with simple, open, and immediate access to all data as well as to derived results. And by so doing, we hope to act as a facilitator of science and to ensure synergies and the maximal utility of these data are obtained. So given that we need this, and I think it's fairly obvious that we need it, what sort of principles should this data coordination platform follow? It should be built from modular components that interact via standards and protocols rather than a monolithic system, and it should be open access. It should be open source. People should be able to adapt it. People should be able to contribute to it. By doing that, it will be flexible and it will be generalizable. It should also be constructed so as to be allowing extensions, and it should be the bedrock for applying the multiple beautiful analysis strategies that we've heard about earlier this morning to allow different biological questions to be addressed, because we should always keep in mind that there's no point generating a beautiful computational and software tool if it doesn't facilitate the biology. The point of this is not the generation of data; it is the use of the data to answer interesting biological questions. And these can be around the areas of hypothesis generation via clustering, throughout the visualization techniques we've heard, and ultimately by integrating imaging and sequencing data, which we've also heard about. So given that we need this, and these are the underlying principles, What does this actual pipe process look like, at least in our heads? Well, it's illustrated by this diagram here. Those of you who have worked on this over many months will have this ingrained upon your minds, I think. For those who are less familiar with it, it's probably worth just talking you through it briefly. So we anticipate the data generators here on the left-hand side. They'll generate their data. In the short term, we anticipate this being single-cell RNA sequencing. In the longer term, This will also stretch to imaging data, proteomics data, and also other modalities that are hard to anticipate because the technology moves so fast. Having generated these data, they will be uploaded via multiple different brokers. The different brokers could be catered to different types of technology or cell types. You might have an immune cell broker, you might have a blood cell broker, and— but the point is that these brokers will act as the facilitators via an ingest API to upload the data to this ingest portal. And within this ingest portal, basic quality assurance metadata checks will be made. And it's very important that we do this because we need to have a basic standard for all of the data that's generated in this atlas, and it has to be coherently done, and those checks will take place here. Having been ingested, the data will then be transferred via this producer API into replicated object stores. These will be cloud-provided, and also there will be a disk archive. From there, we've got the data here, the raw data. What can you do with it? It will exit— exit, excuse me— this blue box. And first, if we focus on the top and these green boxes up at the top, it will be processed by different pipelines. These pipelines will be community-developed. They will be motivated by the hackathons that Dana talked about earlier, and they will allow the data to be processed. So for example, with single-cell RNA sequencing, the alignment, the generation of a gene count table. Basic analyses will be performed. These derived results will then be fed via the APIs, the ingest API, into the object store, and then they, together with the raw data, will be set and shared with the entire community via the consumer API. And that will be done via these different portals here. And we anticipate that these portals will be motivated by different tasks. They will be We hope that 1,000 flowers will bloom here, that really there will be multiple different portals and that they will be able to be used for different approaches. And we think that this is very exciting and really will engage the whole community. And the purpose of this on the left-hand side really is to make that data accessible to everybody so that they can generate these portals and thence use these data to analyze and actually address important biological questions. So this is our plan. How are we going to make sure that it's done in an appropriate way? Well, we've got a very important community governance structure. So we have a steering committee which is comprised of myself, Ben, Anthony, and Jeremy, which works together with members of the HCA overall organizing governance committee chaired by Sarah and Aviv. And here we have oversight about what's going on within this project, and we make sure that it's on track. Of course, on a more day-to-day basis, there are weekly calls where different people are in charge of organizing the ingestion, data store, and analysis engineering teams. Together with these teams, these more engineering software-focused teams, we have working groups which focus more on some of the computational methods development in terms of algorithms. And so with the analysis working group, which I chair along with Dana, we're going to interact very closely with the pipelines to ensure that appropriate methods are applied, and there's going to be new metadata standards working groups which will interface with the community to make sure that we upload appropriate data so that the downstream analyses can be performed robustly. And of course, at all levels, via Slack, via other strategies, we are incredibly eager for community engagement and involvement. So where are we at the moment? Well, actually, we've made very, very good progress in the last 6 months, I think. We have focused particularly at various meetings on defining APIs, metadata standards, as well as implement— implementation of prototypes. And via these meetings, which are both in person and via video conferences, we've made really good progress in pushing forward on the data standards and quality metrics. We have a, I think, probably a beta or maybe a gamma version of the analysis pipeline. We hope to develop an end-to-end prototype of the platform by the end of this year, which will be suitable for beta testing, and we anticipate that early next year we'll have an alpha version of this pipeline, at least for single-cell RNA sequencing data, that will be accessible to the whole community and thereby allowing the data to be shared across the whole scientific community.