CellPredX, a computational framework for cross-data type, cross-sample, and cross-protocol cell type annotation...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate cell type annotation is crucial for single-cell analysis but challenging across diverse datasets and modalities.
• Transferring labels between single-cell RNA sequencing (scRNA-seq) and single-cell Assay for Transposase-Accessible Chromatin sequencing (scATAC-seq) data is difficult due to differing protocols and feature spaces.
• Current methods often address only specific challenges, require manual tuning, and lack interpretability.

Purpose of the Study:

• To develop a unified, semi-supervised framework (CellPredX) for accurate and interpretable cross-modality cell type label transfer.
• To address challenges in transferring labels between scRNA-seq, scATAC-seq, and cross-protocol datasets.
• To provide a scalable solution for single-cell multi-omic integration.

Main Methods:

• CellPredX employs a structurally unified yet adaptively parameterized framework with semi-supervised learning.
• It integrates domain adaptation and deep metric learning to align heterogeneous data embeddings.
• A sparse center loss with an attention mechanism enhances discriminative power, and an interpreter module provides biological insights via gradient attribution.

Main Results:

• CellPredX consistently outperformed state-of-the-art methods in accuracy and robustness across scRNA-seq to scATAC-seq, scATAC-seq to scATAC-seq, and scRNA-seq to scRNA-seq transfer tasks.
• The framework demonstrated adaptive hyperparameter tuning for varying dataset similarities.
• The interpreter module identified biologically relevant marker patterns consistent with known cell hierarchies.

Conclusions:

• CellPredX offers a robust, interpretable, and scalable solution for cross-modality cell type annotation.
• The framework advances single-cell multi-omic integration by enabling accurate label transfer.
• CellPredX facilitates deeper biological understanding through interpretable marker identification.