swCAM: estimation of subtype-specific expressions in individual samples with unsupervised sample-wise deconvolution

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Complex tissues contain diverse cell subtypes with varying expression patterns.
• Current computational deconvolution methods provide only population-averaged expression, limiting subtype-specific analyses.
• Inferring co-expression networks within specific cell subtypes from bulk data is challenging.

Purpose of the Study:

• To develop a novel computational method for estimating cell subtype proportions and subtype-specific (STS) gene expressions in individual samples from bulk tissue transcriptomes.
• To enable downstream analyses, such as subtype-specific co-expression network inference, that are not possible with existing methods.

Main Methods:

• Developed sample-wise Convex Analysis of Mixtures (swCAM), an extension of the CAM framework.
• Formulated swCAM as a nuclear-norm and ℓ2,1-norm regularized matrix factorization problem.
• Utilized cross-validation for hyperparameter tuning and an alternating direction method of multipliers for solution computation.

Main Results:

• swCAM accurately estimates STS expressions in individual samples from simulated data.
• Successfully extracted cell subtype co-expression networks previously unobtainable from bulk data.
• Applied to bulk RNASeq data from brain tissues, swCAM identified significant changes in cell proportions, expression patterns, and co-expression modules in neurons of patients with bipolar disorder or Alzheimer's disease.

Conclusions:

• swCAM provides accurate, sample-level deconvolution of transcriptomic data, overcoming limitations of existing methods.
• The method facilitates novel insights into cellular heterogeneity and disease mechanisms at the subtype level.
• swCAM analysis revealed disease-specific alterations in neuronal subtypes in neurological disorders.