Global tree encoding of atlas-scale single-cell genomics

The rapid growth of single-cell genomic data has created the need for efficient and interpretable ways to represent biological information. This is crucial for understanding the complex relationships between different cellular states at various scales, from individual tissues to multi-scale organization. Current methods, which use discrete abstractions, are good at tasks like cell annotation and trajectory inference, but they can sometimes miss the bigger picture of how cells relate to each other and how their relationships change at different levels of detail.

As datasets grow larger, methods that reduce information, like random sampling, can limit the resolution of rare cell types and diverse cellular states. To address these challenges, researchers have developed MILK, a new computational framework designed to handle high-dimensional single-cell data. This system organizes cell populations into structured tree representations, allowing for efficient data management and analysis.

By using MILK, scientists can extract representative samples from large datasets while keeping important information intact. This ability is particularly useful for running complex algorithms such as deep generative models and for benchmarking new models. Additionally, MILK supports comprehensive analyses across different levels of resolution, including the study of developmental pathways, the impact of diseases on cells in various tissues, and the comparison of genetic programs across different species.

By presenting biological data in a hierarchical way, MILK offers a scalable and flexible method for analyzing single-cell genomic information, making it easier to integrate data from different studies and experiments.