Pasta: open aging clocks help select the next cell experiment
Pasta: open aging clocks help select the next cell experiment
On July 27, researchers at Karolinska Institutet published a paper on Pasta, an open software suite for transcriptome analysis. It estimates age-related shifts from gene activity and suggests which chemical or genetic perturbations should be tested in cells.
In the paper, Jérôme Salignon and colleagues trained Pasta on 17 212 samples from healthy people across 21 studies. The transcriptome is the set of genes that a cell is using at a given time. The model ranks genes within each sample by their relative activity and uses this ranking to calculate a relative age score. A single model works with several methods for measuring gene activity: conventional RNA sequencing, single-cell data, and older microarrays.
Aging clocks usually provide only a number that indicates whether a sample appears younger or older. However, that number depends on the model. Eight epigenetic clocks assigned age estimates to the same blood samples that differed by an average of 17 years. The Pasta authors found another use for the age score. They applied the model to the Connectivity Map, a repository of cellular responses to compounds and genetic changes. It contains more than three million transcriptomes from 248 cell lines exposed to more than 30 thousand chemical and 14 thousand genetic perturbations. For each perturbation, Pasta compared treated cells with control cells from the same experimental batch and ranked candidates for the next experiment.
The screen identified 271 chemical perturbations associated with a higher age score and 63 associated with a lower score. The authors tested two predictions in cells. The drug pralatrexate induced signs of cellular senescence in the A375 melanoma cell line: the cells became larger, divided more slowly, and showed higher levels of p21 and IL6. In the PC3 prostate cancer cell line, piperlongumine increased the mRNA levels of OCT4, SOX2, and NANOG, which are genes associated with a stem-like state. The predictions were not reproduced in the preselected comparison cell lines. Pralatrexate did not induce the key signs of cellular senescence in MDA-MB-231 cells, while piperlongumine did not produce the same set of markers in MCF7 cells.
Pasta’s main output is not a label for a molecule, but a specific, testable hypothesis: which perturbation to test in which cellular context. The same molecule can alter gene activity differently in different cells. This type of screen therefore provides a map for subsequent experiments rather than a universal answer. The open-source code and pretrained models allow other groups to test these candidates in their own cells and tissues.