Fast and interpretable consensus clustering via minipatch learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Consensus clustering improves accuracy and stability by aggregating results from multiple runs.
• Current methods struggle with large-scale bioinformatics data due to computational inefficiency and lack of interpretability.
• Discovering cell types from single-cell sequencing data highlights these limitations.

Purpose of the Study:

• To address the computational inefficiency and lack of interpretability in consensus clustering for large-scale bioinformatics data.
• To develop a novel consensus clustering approach that is both efficient and interpretable.
• To enable more reliable analysis of complex biological datasets, such as single-cell sequencing data.

Main Methods:

• Developed IMPACC (Interpretable MiniPatch Adaptive Consensus Clustering), a novel consensus clustering algorithm.
• Ensembles cluster co-occurrences from minipatches (tiny subsets of observations and features) to reduce computation.
• Incorporates adaptive sampling schemes for observations and features to enhance reliability, efficiency, and interpretability.

Main Results:

• IMPACC significantly reduces computation time compared to standard consensus clustering.
• The method yields more accurate and interpretable clustering solutions.
• Adaptive feature sampling quickly identifies key features differentiating clusters.

Conclusions:

• IMPACC offers a computationally efficient and interpretable alternative for consensus clustering in bioinformatics.
• The approach is effective for large-scale datasets, including single-cell sequencing data.
• IMPACC advances the analysis of complex biological data by providing more reliable and understandable clustering results.