Recursion, Roche, and its Genentech subsidiary have unveiled the latest product of their nearly four-year artificial intelligence (AI) drug discovery collaboration, a whole-genome map of specialized microglial immune cells that the companies plan to use toward revealing significant new targets in neurodegenerative diseases.
The microglia map—which the companies have dubbed a “Google Map of the Brain” and call it a first of its kind—consists of 46 million microglial cell images from a massive perturbation set. The map is designed to offer a new tool for investigating the function of the brain’s specialized microglial immune cells, with the goal of advancing discovery and development of multiple new potential treatments for a variety of neurological diseases, including Alzheimer’s disease, Parkinson’s disease, multiple sclerosis, and ALS.
Roche and Genentech have accepted the microglia map, triggering a $30 million milestone payment to them by Recursion. That latest payment raises to $213 million the total cash paid out by the biopharma giants to Recursion since the AI-based drug developer joined with the pharma giant and its subsidiary to launch their up-to-$12 billion partnership in 2021.
Roche/Genentech accounts for nearly half of the more than $500 million in milestone and upfront payments Recursion has garnered across all its partnerships and collaborations. The microglia map is the second neuroscience map to be optioned in the collaboration, and the sixth phenomap overall.
The first phenomap consisted of over one trillion iPSC-derived neural cells, alongside around 5,000 transcriptomes representing approximately 171 TB of data. Four additional maps—all whole genome scale and small molecule GI-oncology specific phenomaps envisioned by the companies—have been delivered as part of the GI oncology portion of the collaboration, Recursion said in reporting second quarter results. The company plans to release Q3 results on November 5.
Roche and Genentech have committed to using Recursion’s namesake Recursion Operating System (OS) to advance therapies in 40 programs that include “key areas” of neuroscience and an undisclosed gastrointestinal (GI)-oncology indication.
Chris Gibson, PhD, Recursion’s co-founder and CEO
“The Recursion Microglia map is a predictive, functional map designed to use AI to find new therapeutic targets by linking the entire genome to cellular behavior in a highly controlled, standardized system,” Chris Gibson, PhD, Recursion’s co-founder and CEO, told GEN. “By observing the relationship between thousands of genes and chemical compounds on microglial morphology and function, their phenotypic readout, we gain a foundational understanding of the causal role of genes in core microglial biology.”
Overcoming discovery hurdles
Gibson said the microglia map was needed to allow Roche, Genentech, and Recursion to overcome two major hurdles in conventional neuroscience drug discovery: bias toward failed hypotheses and the inefficiency of low-throughput methods.
Traditional drug discovery for neurodegenerative diseases has been heavily biased toward a few well-studied pathways—a “lamppost” effect that has led to high failure rates, since the underlying disease biology is far more complex and more poorly understood than accounted for in any single target-based hypothesis. Neuro drug candidates fail about 85% of Phase II and III trials, according to the global, precision medicine clinical research organization Precision for Medicine.
To address that shortcoming, Recursion said, the microglia map provides an unbiased, holistic view of the entire network of biology. By performing a whole-genome screen on more than 17,000 genes, the map lets researchers explore the unexplored biology around known “bright spots,” driving the discovery of completely novel genes and pathways for first-in-class therapies.
The map is also designed to avoid traditional drug discovery inefficiency by offering a dataset large enough to let machine learning (ML) models systematically evaluate thousands of gene targets at once. The models aim to discover subtle, high-dimensional relationships between genetic changes, chemical compounds, and resulting cell phenotypes that humans would likely miss.
Conventional discovery, on the other hand, involves scientists manually proposing and validating potential targets based on literature, usually one at a time—a process slow and inefficient enough to subject research to human biases.
The companies applied a large-scale discovery approach in their first whole-genome “Neuromap” focused on neuronal cells. That Neuromap, which Roche and Genentech accepted last year, has already identified biological insights that hold potential to become novel targets of interest. The microglia map aims to replicate and expand that success in the brain’s resident immune cells.
“Given the growing evidence linking microglial dysfunction to chronic inflammation and tissue damage across the nervous system, we anticipate implications for treating a wide range of neuroinflammatory and neurodegenerative diseases,” Gibson said.
Differences with HuMicA
The microglia map differs fundamentally from single-cell microglia atlases like the Human Microglia Atlas (HuMicA), published in January in Nature Communications, and older maps. Instead of simply describing cell states, the microglia map predicts gene function on a massive, standardized scale.
The key differences between the microglia map and HuMicA, Gibson and Recursion explained, are how data is generated and what the map represents.
One difference centers around data. The microglia map was created by actively perturbing more than 17,000 genes through genome-wide knockdown/overexpression involving approximately 100,000 sgRNA CRISPR-Cas9 knockouts, adding thousands of chemical compounds, and other perturbations in a massive-scale, standardized batch of cells, then recording the visible phenotypic consequence.
Recursion, Roche, and Genentech reason that their approach to data offers the benefit of going beyond observing what factors are correlated with disease, instead attempting to uncover what causes or rescues a specific dysfunctional state, a key for identifying drug targets.
HuMicA, by contrast, integrates real human disease data to observe the microglial transcriptional landscape, showing what cell subpopulations exist, such as Inflam.DAM (inflammatory disease-associated microglia) and Lipo.DAM (DAM from a subpopulation of cells within adipose tissue), and how their proportions change across diseases ranging from Alzheimer’s disease to multiple sclerosis to COVID-19. That approach, however, is limited by the heterogeneity and low yield of postmortem tissue data, Recursion, Roche, and Genentech assert.
The other key difference between the microglia map and HuMicA centers around scale. The microglia map was created from more than 100 billion human induced pluripotent stem cell (hiPSC)-derived microglia, captured in 46 million images, a scale the companies call critical to generating high-quality data reliable enough for AI/ML models. HuMicA integrates 90,716 cells/nuclei, reflecting multiple studies that together present inherent batch effects and a lack of uniformity.
Tech applications
Building the microglia map required Recursion, Roche, and Genentech to apply technologies perfected in recent years, including:
Differentiation of human induced pluripotent stem cells (hiPSCs) into functional, human microglia.
Massive-scale manufacturing of over 100 billion hiPSCs in a standardized, reproducible way, to enable high-throughput perturbation and analysis.
High-throughput phenotypic screening designed to observe complex changes in cell behavior and morphology rather than focus on a single protein target—an unbiased approach that the companies say captures system-wide effects relevant to complex diseases.
Automation: Recursion used its high-throughput screening (HTS) platforms, which automated the testing of 17,000 genetic perturbations and thousands of compounds, to generate raw data efficiently.
High-dimensional feature extraction intended to capture the cellular effects of gene perturbation in 46 million images. Advanced deep learning models analyzed the images, extracting vast numbers of high-dimensional phenotypic features otherwise invisible to the human eye.
Functional mapping: Recursion’s AI models converted the data into a navigable digital map designed to predict the relationship between any pair of genes or compounds.
To create the map’s 46 million microglial cell images, advanced cell manufacturing protocols were used to grow >100 billion standardized, hiPSC-derived microglia in the lab. The cells were systematically perturbed using genetic editing tools such as CRISPR-Cas9 to knockdown or overexpress over 17,000 genes, or by adding chemical compounds.
High-powered, high-throughput microscopes then automatically captured the resulting changes in the cell’s appearance and structure, generating the vast volume of image data seen in the map. The data was fed into ML models that analyze the subtle, high-dimensional features, effectively completing the process of building the navigable digital map.
Gibson said the map has been designed to enable potential expansion into multi-modal maps by layering on additional, complementary data, such as single-cell screens and transcriptomics data, which measure gene expression.
The post Recursion, Roche Unveil Microglia Map of Neuro Disease Targets appeared first on GEN – Genetic Engineering and Biotechnology News.
