BOLTZMANN LEARNING CODE

Code dca.cpp (c++ language) implements a parallelized sthocastic gradient descent procedure adopted to solve the direct coupling analysis inverse Potts problem on RNA families. It should be compiled with a C++11 supporting compiler. Recent GCC versions support C++11 by default, whereas older version require -std=c++11 or -std=c++0x flag.

EXECUTION

Script run.sh must be executed. It requires two arguments:

• first argument must be the input file (any name), a multiple sequence alignment of homolougus RNA sequences in FASTA format (alphabeth AUCGaucg-)
• second argument must be the number of processes to be run in parallel The script takes care of proper manipulation of the input file and of compiling/executing the main code dca.cpp in parallel (MPI protocol)

EXAMPLE

./run.sh MSA 20

where MSA is a reported input file example and 20 would be the required number of processors

Notice that the run.sh script preprocess the MSA first and produces a file named seqs with only the columns corresponding to the non-gap positions in the first sequence of the alignment. An example seqs.test file already preprocessed is also included in this repository. It can be analyzed using the command

mv seqs.test seqs
mpirun -np 20 dca.o 

OUTPUT

• conv_fi, conv_fij: empirical frequency counts (one-site and two-sites respectively) vs model frequency counts (obtained via Monte Carlo sampling from the obtained Potts distribution)
• h_vals, j_vals: all parameters resulting from learning procedure (local fields and direct couplings respectively)
• scores: pairs covariance scores and respective nucleotides indexes (couplings Frobenius norm and subsequent APC correction)