Multimodal Analysis of White Blood Cell Differentiation in Acute Myeloid Leukemia Patients using a β-Variational Autoencoder

TL;DR

The paper addresses how to integrate morphology from single-cell images with transcriptomic profiles to understand white blood cell differentiation in AML. It introduces a beta-VAE–based multimodal framework with a Mask R-CNN prior to isolate single cells, mapping image features and scRNA-seq into a shared latent space without supervision. The approach enables joint reconstruction and smooth trajectory interpolation between progenitor and mature cell types via Gaussian Process regression, revealing coupling between cellular morphology and gene expression. The results rediscover known driver genes and propose additional candidates, demonstrating potential for improved diagnostics and mechanistic insight, while noting data limitations for trajectories and the need for further validation.

Abstract

Biomedical imaging and RNA sequencing with single-cell resolution improves our understanding of white blood cell diseases like leukemia. By combining morphological and transcriptomic data, we can gain insights into cellular functions and trajectoriess involved in blood cell differentiation. However, existing methodologies struggle with integrating morphological and transcriptomic data, leaving a significant research gap in comprehensively understanding the dynamics of cell differentiation. Here, we introduce an unsupervised method that explores and reconstructs these two modalities and uncovers the relationship between different subtypes of white blood cells from human peripheral blood smears in terms of morphology and their corresponding transcriptome. Our method is based on a beta-variational autoencoder (ß-VAE) with a customized loss function, incorporating a R-CNN architecture to distinguish single-cell from background and to minimize any interference from artifacts. This implementation of ß-VAE shows good reconstruction capability along with continuous latent embeddings, while maintaining clear differentiation between single-cell classes. Our novel approach is especially helpful to uncover the correlation of two latent features in complex biological processes such as formation of granules in the cell (granulopoiesis) with gene expression patterns. It thus provides a unique tool to improve the understanding of white blood cell maturation for biomedicine and diagnostics.

Figures (3)

• Figure 1: Our method is able to jointly embed and reconstruct single-cell image data and transcriptomic information. A) single-cell images are analyzed using a mask R-CNN for feature extraction and are embedded into a joint latent space with encoded single-cell RNA expressions. B) Interpolations defined on the latent space can be decoded into single-cell images and gene expression variation along the trajectory. Expression variations are clustered showing genes categorized based on different transcriptomic kinetics.
• Figure 2: Single-cell image data is embedded with scRNA-seq data into a continuous and morphologically separated latent space (left). Alignment of gene expressions and image data for selected classes demonstrates a close overlap enabling proper reconstruction of both modalities (right).
• Figure 3: Samples from three different trajectories defined in experiments. For every trajectory the generated images, gene expression variation, and gene expression clusters are shown.