adabmDCA

adabmDCA infers maximum-entropy Potts and Ising statistical models from multiple sequence alignments (MSAs) to extract couplings and fields that characterize residue conservation and epistatic coevolution in protein and RNA families.

Key Features:

• Model Inference: Estimates local fields (biases) and pairwise couplings for Potts/Ising variables from MSAs to capture residue conservation and epistasis.
• Three-Dimensional Contact Map Prediction: Uses inferred couplings to predict residue–residue contacts and three-dimensional contact maps of target domains.
• Mutational Effect Assessment and Sequence Generation: Evaluates models for predicting mutational effects and for generating in silico functional sequences.
• Adaptive Learning Framework: Implements Boltzmann machine learning with both equilibrium and out-of-equilibrium learning methods to accommodate different computational constraints.
• Parameter Pruning: Applies an information-based criterion to prune irrelevant parameters and reduce model complexity.

Scientific Applications:

• Protein and RNA Families: Applied to MSAs of protein and RNA families to study conservation and coevolution.
• Domain-Specific Modeling: Used to model the Kunitz and Beta-lactamase2 protein domains and the TPP-riboswitch RNA domain.
• Structural and Evolutionary Studies: Employed for inferring contact maps and generating synthetic sequences relevant to structural biology, evolutionary analysis, and synthetic sequence design.

Methodology:

Boltzmann machine learning with gradient-ascent optimization of the likelihood, model observables computed via Markov Chain Monte Carlo (MCMC) sampling, support for equilibrium and out-of-equilibrium learning, and an information-based parameter pruning criterion.