AI models enable cross-species mapping of cell biology
Researchers at Stanford Medicine have developed two new artificial intelligence models of the biological cell. The first, called universal cell embedding, paved the way for a second-generation model called TranscriptFormer, which has been trained on data from 112 million cells representing 12 species, ranging from single-celled yeast to humans.
Researchers at Stanford Medicine have created two advanced AI models to revolutionize the study of cell biology. The first model, called "universal cell embedding," laid the groundwork for the second model, "TranscriptFormer." TranscriptFormer has been trained on data from 112 million cells belonging to 12 different species, from single-celled yeast to humans.
These AI models enable scientists to more easily compare cells across species, potentially leading to new insights into diseases and novel treatments. The researchers aim to pave the way for a new field of study and a decade of research.
The human genome contains the instructions for life, but modern biology often focuses on gene expression—the genes a cell is actively using. For instance, beta cells in the pancreas express genes for insulin, while B cells produce antibodies. Each cell type has a unique gene expression pattern, which can differentiate healthy cells from diseased ones or cells from various species.
Building on existing cell atlases—databases of gene expression profiles for hundreds of millions of cells from many species—TranscriptFormer uses algorithms to analyze this complex data, much like large language models like ChatGPT. However, instead of predicting missing words, TranscriptFormer predicts missing gene expression values.
The resulting model creates a "universal space" where all cells from every organism can be placed and compared based on their similarity or differences. This shared mathematical space allows scientists to explore evolutionary relationships, such as identifying the closest cell type to a sponge choanocyte or determining the function of neuroid cells in sponges.
TranscriptFormer can also classify cells in new species based on gene expression data it hasn't been trained on. It can distinguish healthy cells from diseased ones, which could aid in designing new cell-based treatments. The researchers hope TranscriptFormer will serve as a powerful tool for biologists to tackle complex problems.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
