Genetics

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

Fenosoa Randrianjatovo, Maya Saleh, Simon Girard, Amadou Barry

Featured September 5, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content.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

Simply

CosMAP creates clearer data maps by first finding neighbors using pattern similarity, then cleaning up these neighbor connections in a hidden step, and finally arranging points so good neighbors stay close and others push apart.

In depth

CosMAP introduces a novel unsupervised dimensionality reduction method that combines cosine-similarity neighborhoods with temperature-normalized contrastive affinities to create faithful low-dimensional embeddings. It further employs a two-phase refinement strategy, first learning an intermediate higher-dimensional representation to denoise the neighborhood graph, and then using this refined graph to optimize the final 2D embedding, thereby improving local and global structure preservation.

Key Takeaways

1
CosMAP leverages cosine similarity for constructing high-dimensional neighborhood graphs, which is particularly effective for sparse omics and image data by focusing on angular relationships rather than magnitude differences.
2
The method integrates contrastive learning principles by defining temperature-scaled affinities for positive pairs and optimizing an attraction-repulsion objective with negative sampling, leading to clearer separation of clusters.
3
A two-phase refinement strategy is introduced, where an intermediate higher-dimensional embedding is learned to reconstruct a more reliable neighborhood graph before the final 2D projection, addressing the issue of noisy initial graphs.

Conceptual Flow

HIGH LEVEL

How CosMAP Creates Clearer Data Maps

CosMAP first finds neighbors using pattern similarity, then cleans up these connections in a hidden step, and finally arranges points so good neighbors stay close and others push apart.

Complex Input Data

Find Patterns & Refine

Simple Data Map

CosMAP's Improved Data Separation

CosMAP consistently creates data maps where different groups are much more clearly separated and organized compared to other common methods.

Messy Data Map (Others)

Mixed Data Map (Others)

Better Grouping

Clear Data Map (CosMAP)

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.