Artificial Intelligence Aids in Providing a Comprehensive View of Cellular Interactions
Understanding the origin of diseases and predicting the success of various treatments often requires studying gene expression in a patient’s cells. This is particularly true for cancer patients. However, obtaining a complete picture of a cell’s state is a complex process due to the intricate and multilayered nature of cells. The data obtained can vary depending on what measurements are taken. For example, the information gained from measuring proteins differs from that obtained by measuring gene expression or cell morphology.
The Challenge of Multimodal Measurement
Obtaining complete information about a cell’s state often necessitates multiple measurements using different techniques, which then need to be individually analyzed. Current machine learning methods can expedite this process, but they aggregate all data from each measurement modality. This makes it difficult to discern which data comes from which part of the cell. Gathering data from different parts of the cell is essential to understand the cell’s state fully.
AI Framework for Better Understanding of Cell State
Researchers from the Broad Institute of MIT and Harvard, and ETH Zurich/Paul Scherrer Institute (PSI), have developed an AI-based framework to address this challenge. This ground-breaking framework learns what information about a cell’s state is shared across different measurement modalities and what data is unique to a particular type of measurement.
By identifying which information comes from which parts of the cell, this approach offers a more holistic view of the cell’s state. This comprehensive view can aid biologists in understanding the complete picture of cellular interactions, which is crucial in understanding disease mechanisms and tracking the progression of diseases like cancer, Alzheimer’s, and diabetes.
Integration of Multiple Measurements for Comprehensive Analysis
The newly developed machine learning framework can differentiate between overlapping information from different modalities and unique information specific to a particular modality. This allows for the automatic identification of shared and modality-specific data, reducing the need for multiple individual experiments and comparisons.
Typical autoencoder machine learning models encode a separate representation of data collected by each measurement modality. The MIT method, on the other hand, has a common representation space for encoding overlapping data from multiple modalities, as well as separate spaces for encoding unique data from each modality. This can be visualized as a Venn diagram of cellular data.
Testing and Future Applications
The AI-based framework was tested on synthetic datasets and was able to accurately capture known common and modality-specific information. When applied to real-world single-cell datasets, it was able to automatically distinguish between gene activity captured jointly by two measurement modalities, and identify information specific to one modality.
The researchers also used their method to identify which measurement modality captured a specific protein marker indicating DNA damage in cancer patients. This information is crucial for clinical scientists in deciding which technique to use for measuring this marker.
The research team aims to further enhance the model’s ability to provide more interpretable information about the cell’s state and to conduct additional experiments to ensure the accurate disentangling of cellular information. They also plan to apply the model to a wider range of clinical questions.
This innovative project was funded by various organizations, including the Eric and Wendy Schmidt Center at the Broad Institute, the Swiss National Science Foundation, the US National Institutes of Health, the US Office of Naval Research, AstraZeneca, the MIT-IBM Watson AI Lab, the MIT J-Clinic for Machine Learning and Health, and received a Simons Investigator Award.
This new AI technology promises to revolutionize the way cells are studied, providing a more comprehensive understanding of cellular interactions and disease progression. For more information about this research, please click here.
