PlmDCA

Pseudo-likelihood maximization in Julia. A complete description of the algorithm can be found at http://plmdca.csc.kth.se/. If you use this algorithm you should cite:

M. Ekeberg, C. Lovkvist, Y. Lan, M. Weigt, E. Aurell, Improved contact prediction in proteins: Using pseudolikelihoods to infer Potts models, Phys. Rev. E 87, 012707 (2013)
M. Ekeberg, T. Hartonen, E. Aurell, Fast pseudolikelihood maximization for direct-coupling analysis of protein structure from many homologous amino-acid sequences, arXiv:1401.4832 (supplementary material)

The present software is a Julia implementation of above mentioned papers, with no reference to the original MATLAB software implementation.

Overview

The code uses NLopt which provides a Julia interfaces to the free/open-source NLopt library. The program (only in its asymmetric version so far, see below) can be run on multiple cores previous addprocs(nprocs) where nprocs should be some integer number np lower or equal to your (physical) number of cores.

Install

It requires the installation of:

NLopt.jl.

julia> Pkg.add("NLopt")

GaussDCA

julia> Pkg.clone("https://github.com/carlobaldassi/GaussDCA.jl")

[PlmDCA] The PlmDCA module itself can be added with

julia> Pkg.clone("https://github.com/pagnani/PlmDCA")

We have not tested yet the software on Windows.

Usage

To load the code just type

julia> using PlmDCA

The software provides two main functions plmdca(filename::String, ...) and plmdcasym(filename::String,...) (resp. the asymmetric and symmetric coupling version of the algorithm). Empirically it turns out that the asymmetric version is faster and more accurate. This function take as input the name of a (possibly zipped) multiple sequence.

We also provide another function mutualinfo(filename::String,...) to compute the mutual information score.

There are a number of possible algorithmic strategies for the optimization problem. As long as local gradient-based optimization is concerned, this is a list of :symbols (associated to the different methods):

:LD_MMA, :LD_SLSQP, :LD_LBFGS, :LD_TNEWTON_PRECOND
:LD_TNEWTON_PRECOND_RESTART, :LD_TNEWTON, :LD_VAR2, :LD_VAR1

After some test we found that the best compromise between accuracy and speed is achieved by the Low Storage BFGS method :LD_LBFGS, which is the default method in the code. The other methods can be set changing the default optional argument (e.g. method=:LD_SLSQP).

There are more optional arguments that can be set (to be documented...).

Output

The functions output a type PlmOut (say X) with 3 fields:

X.Jtensor: the coupling matrix J[ri,rj,i,j] a not symmetric q x q x N x N array, where N is the number of residues in the multiple sequence alignment, and q is the alphabet "size" (typically 21 for proteins).
X.pslike: the pseudolikelihood
X.score: a (Int, Int, Float64) (in julia version 0.3) or a Tuple{Int,Int,Float64} in (in julia version 0.4) vector of Tuples containing the candidate contact descending order (residue1, residue2 , score12).

Requirements

The minimal julia version for using this code is 0.3. It requires the package Compat to guarantee compatibility between different Julia versions.

Todos

A lot!