ExPert infers causal perturbations from cell states across biological and clinical contexts

A team of researchers developed ExPert, a method that learns perturbations from observed cell states across 12 different cellular contexts. Trained on 8,964 perturbations and 19.5 million cells, ExPert ranks 19,895 candidate genes per query, many of which were not present in the training data. The method more than doubled the recall of the best existing method with 1,000 times less computational effort.

ExPert has demonstrated its utility in various biological and clinical contexts. In the case of alcohol-associated liver disease, ExPert identified mRNA export as a potential driver, a finding that remained consistent even when the gene expression levels did not change. The researchers then confirmed their hypothesis through experimental observations, such as nuclear poly(A) retention and isoform switching.

In primary mitochondrial disease, ExPert was able to recover the affected metabolic complex from blood samples, localize the signal to monocytes, and even flag a misdiagnosed control case. This showcases the method's ability to transfer and generalize knowledge across different cell types and perturbation types.

In essence, ExPert enables scientists to query perturbation atlases in reverse, starting from a particular cell state and identifying the most probable driver of that state. This reverse-engineering approach could significantly accelerate research in various fields by allowing for faster and more accurate identification of causative factors.