2024 in Review: How AI Shaped a Year of Impact at CZI

Join us as we look back on 2024. We're excited by the advancements we've made in uniting AI and biomedical research to build towards AI-powered virtual cell models, our work in education to help educators bring learning science to life in classrooms, the impact of our scientific institutes, and the collaborations that shape our work - all driven by a shared commitment to building a better future for everyone.

Leveraging AI To Accelerate Our Work

As advances in AI continue to revolutionize scientific discovery, CZI's work over the years has positioned us to integrate it as a core part of our efforts, driving progress toward our goal of harnessing technology to unlock the mysteries of the cell.

There are a lot of unknowns about the smallest unit of life - the cell - which is why scientists have limited understanding of how cells, tissues and organs become diseased and what it takes for them to be healthy. AI offers unprecedented opportunities to change that.

To support the scientists on the frontlines of this research, CZI built one of the world's largest computing systems for nonprofit life sciences research to enable AI for biology at scale. CZI is the only philanthropic organization that is funding and building a computing system of this kind for the scientific community to build AI-powered virtual cell models - digital tools trained on vast biological datasets. These tools will help researchers explore how cells respond to different conditions. Imagine asking a virtual cell model, "What happens if I change the expression of this one gene?" and getting instant predictions of cellular responses.

We're also opening applications for researchers to use our computing to develop models that cannot be created with conventional university resources. Applying AI to biology has already accelerated many research processes, paving the way for discoveries that can lead to groundbreaking treatments and therapies and, ultimately, CZI's aim to cure, prevent or manage all diseases by the end of the century.

While much work remains ahead, this effort is already laying the foundation for predictive models of cells and cell systems that could unlock major advancements in science and medicine. These models will not only enhance our ability to study biological systems but also foster collaboration between biologists and machine learning (ML) researchers to continue to develop and improve these tools.

We closely collaborated with leaders in science, AI and ML to explore how virtual cell models can transform scientific discovery. This collaboration culminated in a preprint outlining a bold vision and roadmap for a large-scale, collaborative research agenda that was recently published in the journal Cell. The paper emphasizes the need for open science, broad community involvement, and long-term commitment to developing and deploying AI-driven virtual cells.

The vision outlined is starting to take shape with the release of SubCell, the first iteration of an image-based cell model developed by CZI AI resident Emma Lundberg and her team. Trained on 1.1 million cell images from the Human Protein Atlas, SubCell predicts drug treatment effects on proteins, offering insights into how cells respond to interventions.

Another model, developed by CZI's head of AI, Theofanis Karaletsos, is scGenePT, which integrates language data from scientific literature with biological data from lab experiments. This dual-training approach allows scGenePT to predict cellular responses to gene alterations or drug exposure with greater accuracy than models trained on a single data type. Such advancements in modeling single-cell perturbations are critical for understanding how genetic changes or treatments may impact cell behavior.

Anaheim Union High School District has developed, trained teachers on, and implemented new tools that leverage AI and build on research from the University of California, Irvine, funded by CZI.

The promise and potential of AI goes beyond science and medicine. We are in a moment where nearly every industry is reimagining itself through the lens of AI. In education, however, significant gaps remain. Many AI tools are built without enough education-specific context, leading to results that often fall short of addressing classroom needs.

To make AI more effective for education, these systems need high-quality, subject-specific knowledge about curriculum, competencies, and learning science. Equally important is the ability to evaluate AI outputs to ensure they meet the standards of accuracy, rigor, and quality essential for teaching and learning. Achieving this requires tools and frameworks developed in collaboration with educators and researchers.

Our education team is thoughtfully exploring the common AI infrastructure that should exist and be open and accessible to all to enable impactful products that help teachers bring learning science into classroom practice.

Later this month, we'll share more about our work to empower education technology developers to build impactful products that help teachers more easily bring learning science into classroom practice. Specifically, we are focused on helping developers enhance AI system inputs by aligning them with learning science research, state academic standards, and curricula. We also aim to help developers evaluate AI system outputs to ensure they meet the accuracy, rigor, and quality essential for teaching and learning.

Momentum Across our Scientific Research Institutes

The Chan Zuckerberg Biohub Network continues to make exciting strides, with this year marking the first time all four of CZI's scientific institutes are fully operational. This collection of collaborative institutes brings together experts from diverse fields, including cell systems, inflammation, immune cells and imaging.

Aaron Bodansky, a critical care fellow in UCSF's Department of Pediatrics (left), and Joe DeRisi, president of CZ Biohub San Francisco. | Photo by Hannah Kortbawi

At the Chan Zuckerberg Biohub San Francisco, President Joe DeRisi and a team of scientists recently uncovered the cause of MIS-C, a life-threatening inflammatory condition that affects some children after COVID-19 recovery. Using a tool called phage immunoprecipitation sequencing (PhIP-Seq), they discovered that children with MIS-C were producing antibodies and T cells that mistakenly targeted their own immune systems, a phenomenon known as molecular mimicry. Their findings, detailed in Nature, provide the first direct evidence of COVID-19 triggering this autoimmune reaction, opening new avenues for studying similar conditions.

CZ Biohub SF scientists Loïc Royer (left) and Merlin Lange work on a light-sheet microscope designed and built at the Biohub.

In another breakthrough, Merlin Lange, Loïc Royer and other scientists at the San Francisco Biohub unveiled Zebrahub, a cutting-edge cell atlas that maps how a single cell transforms into a complete organism. By combining high-resolution time-lapse videos and gene expression data, Zebrahub maps the intricate journey of cells in zebrafish embryos as they "decide" their roles and work together to form a functioning body. This navigable, interactive map is not only advancing developmental biology but also generating the rich, detailed data needed to power AI initiatives like virtual cell models.

This "organ on a chip" model will help scientists at New York Biohub uncover new details about metastasis. | Photo courtesy of Vunjak-Novaković lab

At the Chan Zuckerberg Biohub New York, researchers are exploring the intersection of bioengineering and immunology, working to enhance human immune cells' abilities to detect and address diseases like ovarian and pancreatic cancers at their earliest stages. One innovative project led by biomedical engineer Gordana Vunjak-Novaković involves developing "organ-on-a-chip" devices - bioengineered models that mimic human organs - to explore how cancer cells evade detection and how researchers might counteract that evasive chemistry.

Postdoctoral fellow Hossein Zargartalebi holds a tiny implantable microdevice that can continuously measure concentrations of inflammatory proteins in living animals in real time. | Photo by Luis Fernando Ayala, courtesy of Northwestern University

This year, the Chan Zuckerberg Biohub Chicago marked its first anniversary and celebrated a significant breakthrough: the development of an implantable device for real-time inflammation monitoring. Detailed in Science, this device uses sensors to continuously track protein levels in living animals. This innovation could revolutionize disease prevention by providing continuous insights into how inflammation impacts health.

In Redwood City, California, scientists and engineers at the Chan Zuckerberg Institute for Advanced Biological Imaging are using cryo-electron tomography (cryoET) to capture 3D images of cells in extraordinary detail. These images, called tomograms, provide valuable insights into the architecture of healthy cells and the changes that may lead to disease. However, while the technology to capture tomograms has improved, the process of analyzing and annotating them remains a significant challenge due to the sheer volume of data.

3D tomogram of a human umbilical vein endothelial cell from Dataset 10176 after membrane annotations by MemBrain. Mitochondrial membranes are colored red, other vesicular membranes are colored blue.

To accelerate this process, the CZ Imaging Institute created the CryoET Data Portal, an open-access data repository that contains about 16,000 annotated cryoET datasets in standardized formats. A new correspondence in Nature Methods gives insight into how scientists and ML researchers created the portal, as well as the goals and next steps supporting it.

The Imaging Institute also launched a challenge on Kaggle, inviting global innovators to develop ML algorithms for automated particle detection, speeding up cryoET annotation and unlocking new possibilities.

Fostering Innovative Collaborations

At CZI, we believe the most meaningful solutions to society's toughest challenges emerge when shaped by diverse perspectives and collective effort. This commitment to collaboration is reflected in several efforts we have initiated.

The Human Cell Atlas is one of CZI's first big bets in science. It is an international community dedicated to creating comprehensive reference maps of all human cells to better understand health and disease. As one of the Human Cell Atlas' first and largest funders, CZI was proud to celebrate the publication of more than 40 peer-reviewed papers in Nature and other Nature Portfolio journals. This collection of papers is a testament to the global grassroots federation of researchers that are motivated to discover more about the 37 trillion cells in our bodies and how they function.

Across our science programs, we're advancing shared progress through partnerships that accelerate discovery and make life sciences research more accessible. Through the Essential Open Source Software for Science RFA, we're supporting critical tools that make research more accessible and scalable. In partnership with The Kavli Foundation and Wellcome Trust, we've strengthened open-source projects by improving usability, enhancing documentation, and fostering community engagement.

Similarly, the Rare As One Project aims to elevate patient communities as central stakeholders in research. Foundational to the program is the Rare As One Network, an incubator-style program to help build the capacities of rare disease organizations and their leaders, optimize their efforts and equip them to sustain themselves and their research efforts. The network began in 2020 with 30 patient-led rare disease organizations and now consists of 94 organizations from around the world. A recent report, developed in partnership with the first 30 groups to join the network, highlights the scientific progress and momentum these groups have inspired.

Families attending the SCN2A Family Conference in September 2023 in Bonn, Germany, organized by SCN2A Germany. | Photo by Max Buchholz

Interdisciplinary research is another hallmark of our work. The 2024 Nobel Prize Laureate in Chemistry David Baker and imaging expert Julia Mahamid are developing a universal genetic label for cryoET, using self-assembling proteins to pinpoint targets in dense cellular environments. This innovation has the potential to transform molecular cell biology by enabling scientists to study proteins with atomic precision.

Priscilla Chan joins community leaders and CZI staff member Ruby Bolaria Shifrin in celebrating five years of the Partnership for the Bay's Future at the Commonwealth Club in San Francisco on March 28, 2024.

We also work closely with local organizations that are activating community-centered solutions to address housing affordability. This spring marked five years of the Partnership for the Bay's Future (PBF), one of the nation's largest private-sector housing funds. Since 2019, PBF has protected over 73,000 tenants and financed more than 4,700 homes, supported the passage of 13 local policies, and engaged over 20,000 community members in housing advocacy. By combining innovative financing with a cross-sector approach, PBF has become a model for addressing the housing crisis that other regions are beginning to replicate.

Looking Ahead

At CZI, we're energized by the possibilities ahead and by the tenacity of our teams and partners in making this work possible. Together and through key investments, we're laying a foundation for discoveries and solutions to build a better future for everyone.

Share