ACTOR

ACTOR analyzes RNA isoform proportions in gene expression datasets using a latent Dirichlet model with Dirichlet Multinomial observations to compare sample isoform proportions against reference panels such as GTEx.

Key Features:

• Latent Dirichlet Model: Uses a latent Dirichlet allocation-style framework to decompose isoform expression patterns into components representing distinct splicing events.
• Dirichlet Multinomial Observations: Incorporates Dirichlet Multinomial modeling to account for overdispersion in RNA sequencing count data.
• Variational Bayes Estimation: Employs a variational Bayes procedure to estimate posterior distributions and produce probabilistic group membership assignments for samples.
• Reference Panel Comparison: Compares sample isoform proportions to large public reference datasets such as GTEx to assess tissue-type alignments.
• R package implementation: Implemented as an R package for computational analysis.
• Gene-level similarity identification: Provides gene-level similarity measures and identifies gene classes with similar alignment to GTEx tissues.

Scientific Applications:

• Cancer Research: Identifying isoform expression changes linked to oncogenic processes.
• Retinal Diseases: Understanding molecular bases of retinal disorders through differential splicing analysis.
• Neurological Disorders: Investigating splicing variations that may contribute to neurological conditions.
• Tissue-type classification and disease-state inference: Assigning tissue-type classifications or disease states based on sample splicing patterns.

Methodology:

Computational methods include a latent Dirichlet allocation-style model with Dirichlet Multinomial observations to model overdispersed RNA-seq counts, variational Bayes estimation for posterior inference and probabilistic group membership, and comparison of inferred isoform proportion profiles to reference panels such as GTEx.