Overview of the CellOT model. bethe distribution of single cells was compared to the untreated control condition (ρc) or one of several perturbed states (ρk, ρI, ρmeter,…). These distributions exist in a high-dimensional space of profiled features. bfor perturbation kwe aim to model it with a function Tk Map untreated cells ρc to the person receiving treatment ρk. cwe are missing a pair of measurements, so we assume that the perturbation is transformed as: ρc into the ρk Under the principle of least effort. In particular, we learn Tk Using optimal transport theory, we directly estimate this distribution mapping as the gradient of the optimal transport double potential ∇ gθ. d, OT maps are learned separately for every perturbation. These maps are fully parameterized, so you can, for example, train CellOT on an initially provided set of samples and predict untreated cells originating from new, never-before-seen samples. . credit: nature method (2023). DOI: 10.1038/s41592-023-01969-x
Experts from ETH Zurich, University of Zurich and University Hospital Zurich have jointly developed an innovative method to use machine learning to predict how individual cells will respond to specific treatments. We can now expect more accurate diagnosis and treatment.
Cancer is caused by cellular changes that lead to the growth of pathogenic tumor cells. To find the most effective combinations and doses of drugs, it is advantageous if doctors can look inside the body, so to speak, and determine how drugs affect individual cells.
An interdisciplinary research team of biomedical and computer scientists from ETH Zurich, the University of Zurich and the University Hospital Zurich is now modeling these cellular changes and drug effects with far greater precision and nuance than ever before. We have developed a machine learning approach that can make predictions. in front.
Understanding individual cell responses
In the fight against cancer, a detailed understanding of how individual cells behave in response to drugs is key. After all, drug therapy should ideally destroy only tumor cells. However, if the effect of a drug is known only as a statistical average of a larger population of cells, analysis of drug effectiveness may indicate that certain tumor cells survive the drug due to their properties or acquired resistance. It may go undetected and the cancer may continue. To spread.
Researchers in Zurich have devised a pioneering approach that recognizes the different responses that individual cells can have to drugs within a larger population. Understanding cellular diversity is critical to developing more effective cancer treatments.
“Diversity within a cell group has a profound effect on a cell’s sensitivity or tolerance to change. Rather than basing our understanding on the average response of a cell group, our method allows cells to “We can accurately describe and even predict how a drug will react to a disturbance,” says Gunnar Rötsch, professor of biomedical informatics at ETH Zurich and University Hospital Zurich. I will explain.
This method can be applied to many cell types
Researchers refer to the molecular reactions in which cells respond to chemical, physical, and genetic influences as perturbations. Such disorders can change the affected cells, leading to, for example, cell death. The effects of certain drugs on cancer cells can also be seen as perturbations.
Understanding which cancer cells respond to drugs and identifying the characteristics of cancer cells that develop resistance to drugs is important for developing new therapeutic approaches and strategies. These new treatments may be more effective at inhibiting cell proliferation or killing pathogenic cells.
Their research was published in the current issue.nature method Along with a research summary of their work, the researchers demonstrate that their method is effective not only on cancer cells, but also on other pathogenic cells, for example in the case of lupus erythematosus. This autoimmune disease is usually accompanied by a red rash and can cause inflammation of the chest, heart, or ribs.
It is now possible to predict the responses of individual cells
Another important innovation that emerges from this research is the ability to make predictions. The Zurich researchers are calling their new machine learning method CellOT. CellOT not only evaluates existing cell measurement data to expand our knowledge of cellular responses to perturbations, but also to understand how individual cells respond to perturbations for which responses have not yet been measured in the laboratory. You can also make predictions.
New methods therefore pave the way for more targeted and personalized treatments. Predictions allow us to predict the effects of perturbations on invisible cells, thus indicating how well a patient’s cells will respond to the drug in question. Comprehensive clinical trials are still needed before this approach can be used in hospitals. Researchers have now demonstrated that this method can provide highly accurate predictions.
Machine learning has made such predictions possible. For CellOT, the researchers use new machine learning algorithms and train them using both data from unperturbed cells and data from cells that change after the perturbation response. Along the way, the algorithm learns how the cell perturbation response occurs, how the reaction progresses, and the possible phenotypes of the altered cell state.
Learning is possible with optimal transportation
ETH computer scientists worked closely with a research group led by Lukas Perkmans, professor of cell systems biology at the University of Zurich. Gabriele Gatto, formerly a postdoctoral researcher in Pelkmans’ lab and currently a senior researcher at the Medical Oncology and Hematology Clinic at Zurich University Hospital, uses a technique called 4i multiplex protein imaging to We measured the changes. “CellOT works particularly well with data acquired with this technology,” he points out Pelkmans. In addition, the researchers obtained single-cell RNA data from public databases.
“Mathematically speaking, our machine learning model is based on the assumption that cells change gradually after a perturbation,” said lead author of the study along with Stefan Stark and Gabriele Gatto. , says Charlotte Bunn, who is working towards a Ph.D. Andreas Kraus, Professor of Computer Science and Director of the ETH AI Center. Bunne’s research area is machine learning, and he explains that “these gradual changes in cell state can be well explained and predicted using the mathematical theory of optimal transport.”
Optimal transport (OT) is the field of mathematics for which ETH mathematics professor Alessio Figari was awarded the 2018 Fields Medal. Over the past four years, optimal transport theory has made significant contributions to explaining cellular perturbation responses.
CellOT is the first approach to predict cellular perturbation responses from new samples using optimal transport and machine learning. “His established OT methods cannot predict out-of-sample or out-of-measurement ranges, but CellOT allows you to do just that,” Bunne says.
For more information:
Charlotte Bunne et al, Learning single cell perturbation responses using neural optimal transport,nature method (2023). DOI: 10.1038/s41592-023-01969-x
Neural optimal transport predicts perturbation responses at the single-cell level.nature method (2023). DOI: 10.1038/s41592-023-01968-y
Quote: It is now possible to predict the effects of drugs on individual cells (October 4, 2023) From https://phys.org/news/2023-10-effect-drugs-individual-cells.html 2023 Retrieved October 4th
This document is subject to copyright. No part may be reproduced without written permission, except in fair dealing for personal study or research purposes. Content is provided for informational purposes only.
