Researchers at the Yale School of Public Health have introduced a new method called scCRAFT that makes single-cell RNA sequencing (scRNA-seq) data more accurate and consistent across different batches of experiments. When scientists use scRNA-seq, they often collect data in separate batches, which can lead to technical differences—also called batch effects—that make it hard to compare results. These batch effects can obscure the true biological differences between cell types.

Overview of the scCRAFT pipeline

a Schematic diagram of scCRAFT’s architecture. The algorithm includes a variational autoencoder at its core, which integrates a batch-ignorant encoder and dual generator-decoders, trained through an Evidence Lower Bound (ELBO) loss. A batch classifier undergoes adversarial training alongside the encoder, utilizing cross-entropy loss. Additionally, the model incorporates triplet loss, calculated within the embedding space Z, guided by labels derived from unsupervised clustering performed on X. b Simplified diagram of the conceptual blueprint of scCRAFT, illustrating the synergistic contribution of the VAE, batch classifier, and triplet loss towards the overarching goals of biological conservation and batch correction. c I and II illustrate how the triplets are chosen based on the low and high-resolution clustering and the mechanism of triplet loss training, respectively. IIII demonstrates the application of scCRAFT on a selected subset from the pancreatic islet dataset, which includes three batches and three cell types. The results compare modeling outcomes with and without the implementation of the triplet loss and show the positions of an example triplet before and after training.

scCRAFT solves this by using a special type of machine learning model called an autoencoder, which helps separate real biological signals from unwanted technical noise. It does this in three ways: first, it checks how well the model can recreate the original data (reconstruction loss); second, it aligns batches using a method called multi-domain adaptation loss; and third, it uses a new dual-resolution triplet loss to keep subtle differences between similar cell types intact, even across different batches.

This combination allows scCRAFT to accurately group cell types without blending them together or misclassifying rare cells. The method was tested on simulated and real datasets and outperformed other popular tools, especially when the batches had unbalanced numbers of cells or rare cell types.

Overall, scCRAFT is a promising tool for researchers looking to get clear and accurate results from scRNA-seq data, no matter how many batches their experiments require.

Availability – The source code and usage tutorial for scDREAMER are freely available at https://github.com/ch2343/scCRAFT/tree/main.