Why do I need to scale CCA components for larger protein panels?

As protein marker counts increase, CCA has more dimensions to fit, increasing the risk of finding 'perfect' but biologically meaningless correlations. Scaling ensures the model captures signal rather than noise.

When should I use a metacell size of 1?

A metacell size of 1 (effectively disabling the feature) is recommended for small protein panels with fewer than 25 markers where the benefit of aggregation is minimal.

What is the sqrt rule for MaxFuse?

The sqrt rule suggests setting the number of CCA components to approximately the square root of the number of active protein features plus one, providing a safe upper bound against overfitting.

How can I tell if my MaxFuse integration is overfitting?

If your canonical correlations remain extremely high (e.g., >0.95) across all components without a smooth decay, the model is likely fitting noise. You should reduce the CCA components.

MaxFuse Parameter Scaling

Name: MaxFuse Parameter Scaling
Author: smith6jt-cop

bysmith6jt-cop

0•

데이터 과학 및 ML

Optimizes MaxFuse integration parameters for high-dimensional protein panels to prevent CCA overfitting during multi-omic data alignment.

This skill provides specialized guidance for scaling MaxFuse parameters when transitioning between different sized protein panels, such as CODEX (26 markers) to PhenoCycler (59+ markers). It implements a robust 'sqrt rule' for Canonical Correlation Analysis (CCA) components and establishes critical Singular Value Decomposition (SVD) limits to prevent noise-fitting in single-cell RNA and protein data integration. By providing verified parameter tables and diagnostic checks, it ensures that biological signals are preserved even when feature dimensions increase significantly relative to cell counts.

주요 기능

01SVD threshold management for RNA and protein features to ensure model stability.

02Standardized parameter scaling tables for 19, 26, and 59+ marker datasets.

030 GitHub stars

04Adaptive metacell size configuration for varying marker counts and panel sizes.

05Diagnostic logic for monitoring canonical correlation decay and distribution alignment.

06Automated CCA component calculation using the sqrt(n_active) rule to prevent overfitting.

사용 사례

01Scaling MaxFuse workflows from low-plex CODEX to high-plex PhenoCycler datasets.

02Calibrating cross-modal data integration when RNA cell counts are limited relative to protein markers.

03Debugging unstable CCA results in single-cell spatial proteomics pipelines.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add smith6jt-cop/skills_registry parameter-scaling

For use in Claude.ai and ChatGPT

Download Skill