MapBatch

Identified Problem

Single-cell sequencing allows for the study of individual cells, but combining data from different sources can be problematic due to batch effects, where variations in data collection can mix things up. Most normalization methods are very aggressive, losing rare cells and removing biological signals along with the batch effects.

AI Solution

MapBatch, a deep-learning tool, normalizes single-cell data by training an autoencoder using a single batch of data to minimize batch effects. It uses an ensemble of autoencoders each trained independently with a single batch to incorporate multiple batches into training. This approach allows for the model to be trained with a diverse set of cells from different batches while avoiding learning batch effects.

Key Features

• Trains an autoencoder using only a single batch of data.
• Uses an ensemble of autoencoders, each trained independently with a single batch.
• Standardizes the data, ensuring that any differences observed are truly significant.
• Preserves rare cell populations that are often lost to other normalization methods.

Benefits of AI Solution

• Helps researchers analyze and compare single-cell data with confidence.
• Identifies rare cell populations associated with disease progression and severity that other normalization tools may obscure.
• Can potentially lead to better prognostic tests and treatments.

Impact of AI Solution

• Enables the discovery of important differences between patients, such as differences in immune cells among patients who relapse from multiple myeloma.
• Facilitates the identification of rare cell populations associated with disease progression and severity in various diseases, including colon cancer and COVID-19.
• Empowers researchers to drive breakthroughs in fields like cancer research and beyond.