climerlab / fbp

Frequency Based Pruning (FBP) is a feature selection algorithm based upon maximizing the Youden J statistic. FBP intelligently enumerates through combinations of features, using the frequency of smaller patterns to prune away large regions of the solution space.

FBP

Frequency Based Pruning (FBP) is a feature selection algorithm based upon maximizing the Youden J statistic. FBP intelligently enumerates through combinations of features, using the frequency of smaller patterns to prune away large regions of the solution space.

Details on FBP 1.0.0 from 2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM).

FBP 1.1.0 updates how the FBP_worker enumerates the patterns.

To Use

Configure the Makefile with the locaion of open mpi libraries and binary

Compile with the Makefile by navigating to the root directory and entering: make

Update configuration file

Run the program. For an example enter: mpirun -np 4 ./fbp <cfg_file>

Configuration

DATA_FILE - Tab seperated file where the first NUM_CASES columns are cases and the next NUM_CTRLS columns are controls. The row indicate features.

SCRATCH_DIR - Directoty where results are recorded

SOL_POOL_FILE - File File with the best and worst objective values from the solution pool of each pattern size.

RUN_TAG - Run tag appended to output file

RISK - Boolean that indicates if risk patterns (true) or protective patterns (false) should be found.

NUM_CASES - The number of cases in DATA_FILE.

NUM_CTRLS - The number of controls in DATA_FILE.

NUM_EXPRS - The number of features in DATA_FILE.

NUM_HEAD_ROWS - The number of header rows in DATA_FILE.

NUM_HEAD_COLS - The number of header columns in DATA_FILE.

PATTERN_SIZE - The number of marker states in the pattern(s) to be found.

MISSING_SYMBOL - String used to indicate missing data in DATA_FILE.

MAX_PS - Maximum size patterns to find

USE_SOLUTION_POOL - A boolean indicating if all pattens above a threshold should be found (true) or just an optimal solution. If true, the worst objective value for each patter size from SOL_POOL_FILE is used as the lower bound.

HIGH_VALUE - Value in DATA_FILE that indicates high expression.

NORM_VALUE - Value in DATA_FILE that indicates normal expression.

LOW_VALUE - Value in DATA_FILE that indicates low expression.

SET_NA_TRUE - Boolean used to indicate if missing data is treated as both high and low.

Outputs

<run_tag>_runtimes.csv - File containing runtimes for each pattern size

PS#_<Risk/Prot>.log - File containing the selected feature sets of size #.

Notes

Recommend using sync-greedy to generate SOL_POOL_FILE.

Requires Open MPI

DATA_FILE should be tab seperate, the columns represent individuals and the rows represent features