PALLAS

PALLAS infers gene regulatory networks from time-series gene expression data using a penalized maximum likelihood framework with LASSO regularization and a continuous-discrete Fish School Search particle swarm algorithm within a Partially-Observable Boolean Dynamical System (POBDS) model to analyze raw continuous microarray and RNA-seq measurements.

Key Features:

• Penalized Maximum Likelihood: Employs a penalized maximum likelihood estimation framework with LASSO regularization to promote sparsity and reduce overfitting in inferred networks.
• Continuous-Discrete Fish School Search Particle Swarm: Uses a continuous-discrete Fish School Search particle swarm algorithm to jointly explore discrete network structures and continuous observational parameters.
• Partially-Observable Boolean Dynamical System (POBDS) Model: Implements the POBDS model to represent gene regulatory dynamics while avoiding ad-hoc binarization of expression data.
• Direct Continuous Data Handling: Processes raw continuous gene expression time-series (e.g., microarray and RNA-seq) without requiring binarization.
• Scalability: Designed to scale to large networks to support comprehensive genomic studies.

Scientific Applications:

• Gene Regulatory Network Reconstruction: Infers GRNs from time-series microarray and RNA-seq expression data for systems biology studies.
• Analysis of Noisy Temporal Expression: Recovers dynamic interactions from noisy time-series datasets to elucidate regulatory dynamics.

Methodology:

Models gene regulation with the POBDS framework to avoid binarization; estimates network parameters via penalized maximum likelihood with LASSO regularization; optimizes jointly over discrete and continuous parameters using a continuous-discrete Fish School Search particle swarm algorithm; validates performance using synthetic datasets (from real and artificial networks) and real-world time-series data.