lmdme

lmdme performs linear model decomposition to denoise and analyze multivariate designed experimental data, leveraging limma's lmFit for ANOVA-style inference and improving detection in low signal-to-noise neuroimaging contexts such as Voltage Sensitive Dye Imaging (VSDI).

Key Features:

• Flexible Formula Interface: Specifies linear models using flexible formula types to encode a wide range of experimental designs.
• Linear ANOVA Decomposition: Implements linear ANOVA-style decomposition using limma's lmFit for multivariate data analysis.
• Computational Efficiency: Leverages limma's efficient computational backend (lmFit) for fast processing of large datasets.
• Statistical Outputs: Provides p-values for estimated coefficients at factor levels and F values for assessing factor effects.
• Spatio-temporal Decomposition: Separates spatio-temporally inseparable noise and signal components and identifies known noise and signal factors.
• Visualization: Generates Principal Component Analysis (PCA) and Partial Least Squares (PLS) plots for multivariate interpretation.

Scientific Applications:

• Enhanced Signal-to-Noise Ratio: Achieves a four-fold improvement in signal-to-noise ratio in reported VSDI analyses.
• Increased Response Detectability: Doubles detectability of neuronal responses by reducing trial-to-trial variability.
• Efficiency in Data Collection: Reduces required trial counts to attain high-quality data, optimizing experimental throughput.
• Versatility in Dynamics Estimation: Accommodates a broad range of response dynamics for estimating spatial activity spread or contrast dynamics.
• Neuroimaging of Trial-Based Experiments: Applicable to neuroimaging studies of trial-based experiments on awake animals, particularly VSDI.

Methodology:

Uses a linear model decomposition inspired by fMRI processing that applies limma's lmFit for ANOVA-style coefficient estimation and inference (p-values and F values), decomposes spatio-temporally inseparable signal and noise while identifying known noise and signal factors, and compares performance against classical denoising methods such as blank division and detrending.