MIT researchers have developed a new theoretical framework to study the mechanisms of treatment interactions. Their approach allows scientists to effectively estimate how treatment combinations will affect a group of units, such as cells, allowing a researcher to perform less expensive experiences while collecting more precise data.
For example, to study how interconnected genes affect the growth of cancer cells, a biologist may need to use a combination of treatments to target several genes at the same time. But because there could be billions of potential combinations for each cycle of the experience, the choice of a subset of combinations to be tested could bias the data generate that their experience generates.
On the other hand, the new framework considers the scenario where the user can effectively design an impartial experience by attributing all the treatments in parallel and control the result by adjusting the speed of each treatment.
The MIT researchers theoretically revealed an almost optimal strategy in this context and have carried out a series of simulations to test it in a multi-terrain experience. Their method has minimized the error rate in each body.
This technique could one day help scientists better understand the mechanisms of the disease and develop new drugs to treat cancer or genetic disorders.
“We've introduced a concept people can think more about as they study the optimal way to select combinatorial treatments at each round of an experiment. Our Hope is this can someday be used to solve biologically raising questions,” Says Graduate Student Jiaqi Zhang, an Eric and Wendy Schmidt Center Fellow and co-lead author of a paper on this experimental design framework.
She is joined on the newspaper by the author of the Co-Divya Shyamamal, a first cycle of the MIT; And the senior author Caroline Uhler, the engineering professor of Andrew and Erna Viterbi in the EECS and the MIT Institute for Data, Systems and Society (IDSS), who is also director of the Eric and Wendy Schmidt center and researcher at the MIT for information and decision -making systems (LIDS). Research has recently been presented at the international conference on automatic learning.
Simultaneous treatments
Treatments can interact with each other in a complex way. For example, a scientist trying to determine if a certain gene contributes to a particular symptom of disease may have to target several genes simultaneously to study the effects.
To do this, scientists use what are called combinatorial disturbances, where they apply several treatments both to the same group of cells.
“The combinatorial disturbances will give you a high -level network of how the different genes interact, which makes it possible to understand the functioning of a cell,” explains Zhang.
Since genetic experiences are expensive and take time, the scientist aims to select the best subset of treatment combinations to test, which is a high challenge due to the large number of possibilities.
Selecting a sub-optimal subset can generate biased results by focusing only on the combinations selected by the user in advance.
MIT researchers addressed this problem differently by examining a probabilistic framework. Instead of focusing on a selected subset, each unit takes treatment combinations according to the dosage levels specified by the user for each treatment.
The user establishes dosage levels according to the objective of their experience – perhaps this scientist wants to study the effects of four different drugs on cell growth. The probabilistic approach generates less biased data because it does not restrict experience to a predetermined sub-assembly of treatments.
Dosage levels are like probabilities and each cell receives a random combination of treatments. If the user defines a high dose, it is more likely that most cells take this treatment. A smaller cell subset will take this treatment if the dosage is low.
“From there, the question is how to design the dosages so that we can estimate the results as precisely as possible?” This is where our theory is coming, ”adds Shyamamal.
Their theoretical framework shows the best way to design these dosages so that we can learn the most about the characteristic or the line they study.
After each round of the experience, the user collects the results and returns those in the experimental framework. He will publish the ideal dosage strategy for the next round, etc., actively adapting the strategy on several laps.
Dose optimization, minimization of errors
Researchers have proven that their theoretical approach generates optimal doses, even when dosage levels are affected by a limited supply of treatments or when noise in the experimental results varies at each turn.
In simulations, this new approach had the lowest error rate when comparing the estimated and real results of versatile experiences, surpassing two basic methods.
In the future, researchers want to improve their experimental framework to consider interference between units and the fact that certain treatments can lead to a selection bias. They would also like to apply this technique in a real experimental framework.
“This is a new approach to a very interesting problem that is difficult to solve. Now, with this new frame in hand, we can think more about the best way to design experiences for many different applications, ”explains Zhang.
This research is funded, in part, by the advanced program of first cycle research opportunities at MIT, Apple, the National Institutes of Health, the Office of Naval Research, the Department of Energy, the Center Eric and Wendy Schmidt at the Broad Institute and a Simons Investigator Award.
