swCAM

swCAM estimates subtype proportions and subtype-specific gene expression in individual samples from bulk tissue transcriptomes to deconvolve cellular heterogeneity and enable sample-level differential expression and co-expression analysis.

Key Features:

• Sample-wise Deconvolution: Provides subtype-specific expression estimates at the individual sample level rather than population-averaged estimates.
• Mathematical Formulation: Extends Convex Analysis of Mixtures (CAM) by incorporating a term for between-sample variations and formulates the problem as a nuclear-norm and ℓ_(2,1)-norm regularized low-rank matrix factorization.
• Hyperparameter Optimization: Determines optimal hyperparameters using a cross-validation scheme with random entry exclusion.
• Efficient Computation: Solves the optimization problem using an alternating direction method of multipliers.
• Differential Expression and Co-expression Networks: Enables extraction of subtype-specific differential expression patterns and co-expression networks within individual samples.

Scientific Applications:

• Accurate Estimation: Demonstrated on realistic simulation data to accurately estimate subtype-specific expressions and recover co-expression networks at the sample level.
• Disease Research (Bipolar Disorder): Applied to bulk-tissue data from 320 samples of bipolar disorder patients and controls, revealing changes in cell proportions, expression patterns, and coexpression modules in patient neurons, including significant alterations in mitochondria-related genes indicative of energy dysregulation.

Methodology:

Unsupervised sample-wise Convex Analysis of Mixtures extended with a between-sample variation term; formulated as a nuclear-norm and ℓ_(2,1)-norm regularized low-rank matrix factorization; hyperparameters selected via cross-validation with random entry exclusion; optimization solved by an alternating direction method of multipliers.