scAClc: A Multi-Objective Adaptive Clustering Framework for Single-Cell Transcriptomics via Contrastive and...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptomic data.
• Accurate cell clustering is crucial for identifying cellular identities.
• Existing clustering methods face challenges like data sparsity, high dimensionality, and noise.

Purpose of the Study:

• To develop a novel clustering framework, scAClc, for scRNA-seq data.
• To address limitations of current clustering methods, including the need for pre-specified cluster numbers.
• To improve the accuracy and interpretability of cell clustering.

Main Methods:

• scAClc employs multiobjective optimization and adaptive resolution discovery.
• Features a Hierarchical Gene Relevance Module for feature selection.
• Incorporates an Anchor-Centered Contrastive Learning Module for embedding.
• Includes a Self-Adaptive Resolution Discovery Module to infer cluster numbers.

Main Results:

• scAClc consistently outperforms six state-of-the-art methods on 15 real scRNA-seq datasets.
• Ablation studies validate the contribution of each module.
• Interpretability analysis enhances understanding of clustering mechanisms.

Conclusions:

• scAClc offers a robust and accurate solution for scRNA-seq data clustering.
• The framework improves the construction of virtual cell representations.
• It facilitates deeper insights into biological mechanisms underlying cell populations.