Adaptive Boltzmann machine learning for Potts models of biological data

Description

This is an implementation of the Boltzmann machine learning to infer several maximum-entropy statistical models of Potts or Ising variables given a set of observables. More precisely, it infers the couplings and the fields of a set of generalized Direct Coupling Analysis (DCA) models given a Multiple Sequence Alignment (MSA) of protein or RNA sequences. It is also possible to infer an Ising model from a set of spin configurations. The learning is performed via a gradient ascent of the likelihood of the data in which the model observables are computed via a Markov Chain Monte Carlo (MCMC) sampling.

First release

The first implementation has been described in adabmDCA: Adaptive Boltzmann machine learning for biological sequences. Please cite this paper if you use (even partially) this code.

adabmDCA has been used in:

• Aligning biological sequences by exploiting residue conservation and coevolution - A. Muntoni, A. Pagnani, M. Weigt, F. Zamponi (on Phys. Rev. E) to learn the Potts model and pseudo Hidden Markov model (see the option -D in Advanced options/Available maximum entropy models) of the studied protein and RNA seed alignment;
• Sparse generative modeling of protein-sequence families - P. Barrat-Charlaix, A. Muntoni, K. Shimagaki, M. Weigt, F. Zamponi (on Phys. Rev. E) to learn the dense model and to perform the information-based pruning of the coupling parameters (see Advanced options/Pruning/activating the coupling).

Last release

Note that the last version of adabmDCA uses a sligthly different set of input flags compared to those presented in adabmDCA. Two learning protocols have been added. It is now possible to:

• Integrate experimental fitnesses in the final model (see Improving landscape inference by integrating heterogeneous data in the inverse Ising problem)
• Perform an edge-activation scheme to train a sparse model. See Towards parsimonious generative modeling of RNA families

Installation

This code is written in C/C++ language. To properly install adabmDCA, run

make

on adabmDCA/src folder. It suffices a g++ compiler.

Usage

All the possible routines (model choices, decimation/activation procedure, input and output files) implemented in adabmDCA can be shown typing

./adabmDCA -h

For a detailed description see the documentation file.