MIT researchers have developed a new theoretical structure to study the mechanism of treatment interaction. Their approach allows scientists to efficiently estimate how the combination of treatment will affect a group of units, such as the cells, such that the cells enable a researcher to make less expensive experiments when collecting more accurate data.
As an example, to study how interconnected gene cancer affect cell development, a biologist may need to use a combination of treatment to target multiple genes simultaneously. But because billions of potential combinations for each round of experimentation can be, choosing a site of combinations for testing can bias the data generated by their use.
In contrast, the new structure considers the scenario where the user can design a fair use efficiently by assigning all the treatments in parallel, and can control the result by adjusting the rate of each treatment.
MIT researchers proved a close-optimal strategy in this structure and demonstrated a series of simulations to test it in a multinational use. Their method reduced the error rate in each example.
This technique can someday help scientists to understand the disease system better and develop new drugs to treat cancer or genetic disorders.
“People can think more about a concept because they study optimal methods to choose a combinatorial treatment in each round of an experiment. We hope that it can be used to solve biologically relevant questions someday,” Graduate students are called Jiaki Zhang, an Eric and Wendy Shamit Center Fellow and a co-worker writer of a paper.
She has joined paper by an MIT undergraduate, co-head-rich writer Divya Shyamal; And senior writer Caroline Uhrar, Andrew and Erna Veterbi EECS Professor of Engineering and MIT Institute for Data, Systems, and Society (IDSS), who are also a researcher in the MIT laboratory for Eric and Weick and Wendy Shamite Center Director and Information and Decision System (LIDS). Research was recently presented at the International Conference on Machine Learning.
Treatment together
Treatment can interact with each other in complex methods. For example, a scientist is trying to determine whether a certain gene contributes to a particular disease symptom, many genes may have to be targeted simultaneously to study the effects.
To do this, scientists are used as a combinatorial disturbances, where they apply multiple remedies in the same group of cells at once.
“Combineatorial disturbances will give you a high-level network of how different genes interact, which gives an understanding of how a cell works,” says Zhang. “
Since genetic experiments are expensive and time -consuming, the purpose of the scientist is to choose the best of treatment combinations for testing, which is a challenge due to the large number of possibilities.
Choosing a subptimal multitude can lead to biased results only by focusing on combinations already chosen users.
MIT researchers approached this problem in a different way by seeing a possible structure. Instead of focusing on a selected most, each unit takes randomly to combine the combination of treatment based on user-specified dosage levels for each treatment.
The user determines the dose level based on the goal of its experiment – perhaps it wants to study the effects of four different medicines on scientific cell development. Possible approach generates less biased data as it does not limit the experiment to the pre -determined mastery of treatment.
The dose level is like possibilities, and each receives a random combination of cell treatment. If the user sets a high dose, it is more likely that most cells will pick up that treatment. If the dose is low then a small satent of cells will pick up that treatment.
“From there, the question is how we design the dose so that we can guess the results as accurately as possible? This is where our principle comes,” Shyamal says.
Their theoretical structure shows the best way to design these doses so that no one can learn about the characteristic or characteristic that they are studying.
After each round of the experiment, the user collects the results and feeds them back into the experimental structure. This will produce the ideal dose strategy for the next round, and so, actively adopt strategy on several tours.
Dosage adaptation, reducing error
Researchers proved that their theoretical approach produces optimal doses, even when the dose level is affected by the limited supply of treatment or when noise in experimental results varies in each round.
In simulation, this new approach had the lowest error rate when comparing the estimated and actual results of multinational experiments, performing better by two baseline methods.
In the future, researchers want to consider their experimental structure to consider the intervention between units and the fact that some treatments may cause selection bias. They would also like to apply this technique in a real experimental setting.
“This is a new approach to a very interesting problem that is difficult to solve. Now, with this new structure in hand, we can think more about the best ways to design experiments for many different applications,” says Zhang.
It is financed by research, MIT, Apple, National Institute of Health, The Office of Naval Research, Energy Department, The Edition of Energy, Eric and Wendy Schm in Broad Institute at the Center, and by a Simmons Investigator Award, and a Simmons Investigator Award.
