Spring 2020 was not a good time to find oneself organizing a scientific meeting. Just as we had begun sketching our outline agenda, the unfolding coronavirus pandemic forced a complete pivot from planning for a regular in-person annual meeting in Midtown Manhattan to reimagining it as an entirely digital event. However, thanks to a wealth of creative thinking and the invaluable can-do spirit of the Simons AV team, together with a lot of leaning-in, the 2020 CBIOMES Annual Meeting turned out to be quite the success, contributing a template for other Simons collaborations to follow, as well as suggesting new online activities that CBIOMES may very well adopt to complement future in-person meetings once we can safely gather again.
CBIOMES seeks to develop and apply quantitative models of the structure and function of marine microbial communities at seasonal and basin scales. In support of this, our 2020 meeting was centered around the overarching and all-important theme of bringing models and data together.
Broadly, day one of the meeting was given over to preparing the ground for break-out sessions planned for day two with the final day reserved for reporting and discussion towards the path ahead. Six topical areas provided focus: Marine Provinces, Bayesian Modeling, Macromolecules and Metabolism, Molecular Data and Biogeochemical Models, Marine Ecosystem Modeling in Julia, and Simons CMAP.
The Biomes Working Group is a cross-institutional collaboration that grew out of discussions at the 2019 annual meeting. Drawing together CBIOMES investigators with expertise in statistical methods, physical and biogeochemical modeling, and remote sensing, the group is exploring questions concerned with spatial and temporal controls on marine phytoplankton distributions observed in nature and in models. Chris Follett (MIT), who informally leads the group, gave an opening introduction to the Marine Provinces sessions, highlighting some of the questions the team has been thinking about, among them questions of monitoring (how centers and boundaries of regions change over time); theories of sparse data (comparing pdf’s of variables in/ across regions over time); algorithm development especially optical algorithms, for Chlorophyll from satellite observations); hypothesis testing (eg why do boundaries/ centers move over time), and model testing (eg just how good is the DARWIN model quantitatively? Can behaviors observed in Darwin help in interpreting data?). On day two, Justin Hyun (USC) seeded discussion by presenting work he has been doing concerning “Ocean Movers Distance”, while Shubha Sathyendranath (PML) spoke on the subject of “Time-varying Provinces.”
The Bayesian Working Group, which grew out of a workshop held at MIT in January 2020, brings together investigators from across the collaboration interested in applying ideas from Bayesian statistics. Day one introductions to the Bayesian Modeling sessions from MIT’s Greg Britten (overview of the group’s activities, available resources), Paul Mattern (UCSC) and UW’s Francois Ribalet (a Bayesian size-structured matrix populations model), and Crispin Mutshunda (Dalhousie) and From Flatiron Aditya Mishra (Dynamic models for multivariate plankton time series) paved the way for day two breakout sessions discussing “Extended modeling frameworks for Seaflow” (led by Francois Ribalet, Paul Mattern, Zhen Wu, and Greg Britten) and “Community dynamics from multivariate time series” (led by Greg Britten, Crispin Mutshinda, Aditya Mishra, and Paul Mattern.)
Julia is a fast, new dynamic programming language with a rapidly growing user community that, within a few short years, has already generated an extensive library of software packages. While it is a general-purpose language many of its features are well-suited for numerical analysis and computational science of the kind used by many within CBIOMES. With the goal of introducing other members of the collaboration to its utility in the Ecological Modeling with Julia sessions, Julia proponents, Gael Forget (MIT) and Zhen Wu (MIT) prepared two ecological modeling examples, one to compute particle trajectories (Gael Forget), the other illustrating agent-based modeling (Zhen Wu). The pair then provided small-group, Jupityr Notebooks-based coaching at break-out sessions on day two.
Metabolic and physiological traits are important factors in determining biogeography but formalizing those relationships requires bridging very different scales. In his introduction to the Macromolecules, and Metabolic Modeling sessions, John Casey (MIT) reviewed the different activities underway in this area within CBIOMES, orienting them relative to one another in terms of biological complexity (eg from simple Monod-based models, up to highly complex genome-scale models) versus diversity (eg from single-cell models to whole ecosystems) setting the stage for day two breakout discussions aimed towards collectively developing a roadmap to guide how to integrate the various threads, and what larger goals to pitch for in the next couple of years.
As explained by Mohammad Ashkezari (UW) in his day one introduction to the CMAP Data Challenge sessions, marine phytoplankton data is messy; collected, and archived in often piecemeal ways. Bringing all that data together in a readily tractable searchable framework for the purposes of intercomparison and analysis has been a key goal of the Simons CMAP atlas, which aims to collect and curate often disparate satellite, in-situ observations and model data, then serve it up in such a way that investigators can devote their time to analysis and not the otherwise time-consuming task of sourcing, preparing and co-locating data. Having provided an overview of CMAP’s architecture and its underlying philosophy, Ashkezari invited those who were interested in putting CMAP through its paces to join him for a hands-on activity on the second day using CMAP to compile and compare picoplankton distributions from the continually expanding cruise data repository with corresponding distributions seen in Darwin.
Finding a common currency by which different kinds of observations can be combined and compared has long been an outstanding question. The Molecular Data and Models sessions introduced by Mick Follows and seeded by presentations from MIT’s Stephanie Dutkiewicz (talking about ecological and biological model simulations and why carbon is an important currency in them), UW’s Ginger Armbrust (talking about efforts to recalibrate cell count data from SeaFlow data into carbon biomass), and USC’s Jed Fuhrman (sharing new ideas and developments in his lab using exact sequence variance data) provided an opportunity for investigators who wrestle with this problem across CBIOMES to come together for a shared discussion exploring how to fuse disparate data into synthetic biogeographies for hypothesis and model testing.
Typically Annual Meeting comprises three back-to-back full-on twelve-hour science days, with just about everyone in the collaboration gathered together at the Foundation to enjoy a perennial high-octane fix of all things CBIOMES: listening science, talking science, even eating science. With this year’s 80+ attendees arrayed over time-zones from Germany to Hawaii, and few with the appetite for three straight twelve-hour days on Zoom even when that mapped to regular local working hours, step one was to limit in-person meeting time to six hours each day (between 11 am and 5 pm EDT daily.) To compensate for this halving of in-person programming time we then opted to significantly reduce the amount of single-presenter plenary presentation in favor of more break-out time, eg arranging for each PI to record a ~10-minute overview of their project to be made available for attendees to review ahead of the meeting.
Sandwiched between early-day and late-day programming, and central to the meeting as a whole, were daily poster sessions. In place of a traditional pdf poster, presenters were asked to share their work via a short deck of 4 or 5 slides. Optional but encouraged was to record a microtalk (<5 minutes) providing narration. All materials were due two weeks before the meeting to allow us to upload them to the virtual gallery built to showcase them. The poster/slide deck and optional microtalks, together with PI project overview videos were then posted for browsing in the week ahead of the meeting. For the “live” poster session during the meeting, poster presenters were encouraged to leverage the novelty of presenting their material online by having the option to give a chalk talk or live Jupyter notebook, or another creative approach as they preferred.