New Computational Approach Enhances Cellular Reprogramming

Scientists from MIT and Harvard University have developed an innovative computational approach to efficiently identify optimal genetic interventions for cellular reprogramming. The novel technique can help reprogram cells to be more effective in fighting diseases like cancer or identify therapies for various disorders.

One significant challenge in cellular reprogramming is the vast number of potential genetic perturbations due to the complexity of the human genome. Conducting experiments to identify the ideal perturbation for a specific application is both time-consuming and costly. Researchers typically struggle to find the most effective genetic interventions due to this challenge.

The MIT and Harvard researchers developed an algorithmic technique that leverages the cause-and-effect relationships within complex systems, like genome regulation. This technique, described in a recent Nature Machine Intelligence paper, helps prioritize the best interventions in a series of experiments efficiently.

The key innovation in their approach is the incorporation of causal relationships into the acquisition function of active learning, a machine-learning approach often used in sequential experimentation settings. While traditional acquisition functions focus on correlation between factors, the researchers' method considers the regulatory relationships or causal structure of the system. For example, it takes into account that a genetic intervention can only affect downstream genes. By utilizing causal models rather than correlation-based models, the search space for interventions becomes smaller, and the acquisition function prioritizes the most informative interventions.

The researchers also introduced an enhancement to their acquisition function using output weighting, inspired by studying extreme events in complex systems. This enhancement focuses on interventions that are likely closer to the optimal intervention, further improving efficiency.

They tested their algorithms using real biological data in a simulated cellular reprogramming experiment. Their acquisition functions consistently outperformed baseline methods, identifying better interventions throughout the multi-stage experiment.

The researchers are now collaborating with experimentalists to apply their technique to cellular reprogramming in the laboratory. Beyond genomics, their approach can potentially address optimization problems in various fields, such as determining optimal prices for consumer products or enabling optimal feedback control in fluid mechanics.

In the future, they aim to expand their technique to address more complex optimization problems and explore how AI can assist in learning causal relationships within systems.