l1kdeconv

l1kdeconv performs deconvolution of LINCS L1000 paired-channel measurements to infer individual landmark gene expression values for downstream analyses.

Key Features:

• LINCS L1000 paired-channel support: Handles the L1000 design where approximately 1,000 landmark genes are measured across ~500 color channels with two genes sharing a channel.
• Outlier detection and removal: Identifies and removes outliers to enhance data quality and stability.
• Aggregate Gaussian Mixture Model (AGMM): Uses an AGMM to separate mixed signals from shared channels into individual gene expression estimates.
• Information borrowing across samples: Leverages data from similar samples to improve deconvolution reliability.
• Benchmarking on simulated and real datasets: Demonstrated improved stability and accuracy relative to commonly used methods in evaluations on simulated and real L1000 data.

Scientific Applications:

• Deconvolution of L1000 measurements: Infers individual landmark gene expression values from shared-channel L1000 data for use in downstream gene-expression analyses.
• Preprocessing to improve downstream analyses: Reduces deconvolution-induced errors to increase reliability of downstream analyses that depend on precise high-throughput gene expression measurements.

Methodology:

Identifies and removes outliers, applies an aggregate Gaussian mixture model (AGMM) to deconvolve paired-channel signals, and borrows information across similar samples.