Genetics

CosMAP: Contrastive Manifold Approximation and Projection for Dimensionality Reduction of Omics and Genealogical Data

Fenosoa Randrianjatovo, Maya Saleh, Simon Girard, Amadou Barry

Featured August 18, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

By using cosine similarity to build better neighborhood maps and a clever two-step denoising process, CosMAP creates clearer, more organized visual groupings of complex data, like cells or family trees.

In depth

The paper introduces Contrastive Manifold Approximation and Projection (CosMAP), a novel dimensionality reduction method. It enhances existing graph-based techniques by integrating cosine-similarity neighborhoods with temperature-normalized contrastive affinities. A key innovation is a two-phase refinement strategy that first learns an intermediate higher-dimensional representation to denoise the neighborhood graph before optimizing the final low-dimensional embedding, leading to more faithful and interpretable visualizations.

Key Takeaways

1
CosMAP improves dimensionality reduction by combining cosine similarity for neighborhood graph construction with contrastive learning principles, which helps preserve local and global data structures.
2
The method employs a two-phase refinement strategy where an intermediate embedding is learned to reconstruct a more reliable neighborhood graph, reducing the impact of noisy high-dimensional connections.
3
Evaluations on omics, genealogical, and image datasets demonstrate that CosMAP produces more coherent and interpretable visual representations with clearer separation of populations compared to state-of-the-art methods.

Conceptual Flow

HIGH LEVEL

Methodology: How CosMAP Works

CosMAP first builds a smart map of connections, then cleans it up in a hidden space, and finally draws a clear picture by pulling similar things together and pushing different things apart.

Messy High-D Data

Build Smart Map

Initial Connection Map

Results: Clearer Data Organization

Compared to older methods that mix things up, CosMAP makes distinct groups stand out clearly, like separating different types of toys into their own boxes.

Mixed Data Points

Old Method

Overlapping Groups

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.