Longevity InTime: Autonomous AI Institute. Anti-Aging Digital Health Immortality Transhumanist AI Channel

PerturbLDM Model The PerturbLDM model predicts how drugs change gene activity in cells under unmeasured conditions. A preprint about PerturbLDM was released on August 12 on bioRxiv. It was tested on combinations of drug, dose, and cell line that were hidden during training, but each drug, dose, and line was separately present in the data. One and the same drug can change gene activity differently in different cell lines - populations of cells that are grown and studied separately. There are so many combinations of drug, dose, and line that laboratories can only measure a part of them. The model is needed to predict in advance which genes will increase or decrease activity in an unmeasured condition. In the preprint, PerturbLDM receives the name of the drug, dose, and cell profile of the same line in a control experiment with DMSO - a solvent that is added instead of the drug. The model first compresses information about the activity of thousands of genes into a compact record, then gradually builds the response taking into account the drug, dose, and control profile, and then translates it back into gene activities. The authors took 46,471 conditions from Tahoe-100M - a large set of single-cell measurements. For training, they left 32,529, and 13,942 were hidden. Both parts included all 379 drugs, 47 cell lines, and three dose levels. Only a specific triplet was hidden: each of its drug, dose, and cell line was separately familiar to the model. 🔗 Read original →