MotifDiscovery

Production-ready Motif GA workflow with optional clustering. Legacy experiments and MATLAB/C NSGA-II code are separated under legacy.

Quick Start (MotifGAImproved)

Prerequisites:

• Java 8+
• Gradle

Build:

./gradlew build

Run (production entry point):

./gradlew run

Optional clustering run:

./gradlew runClustering

Tests:

./gradlew test

GA-Based Dijkstra (Learning)

Run the GA-based shortest path search:

./gradlew runGADijkstra --args="0 3"

Output includes:

• GA best cost/path (heuristic result)
• Optimal cost/path (computed with classic Dijkstra for comparison)

How to learn it:

• GA searches over permutations of nodes (chromosomes).
• Fitness is inverse of path cost; invalid paths get zero fitness.
• Selection + crossover + mutation evolve better paths over generations.
• Compare GA output to the optimal Dijkstra output to validate.

Custom input/output paths:

./gradlew run --args="/path/to/input.txt /path/to/output.txt"
./gradlew run -Dmotif.input=/path/to/input.txt -Dmotif.output=/path/to/output.txt

Non-interactive run (no prompts):

./gradlew run -Dmotif.length=6 -Dmotif.promoter=ATGCAT -Dmotif.parents.percent=50 -Dmotif.motifs.per.sequence=4 \
  --args="data/motifs-input.txt data/motifs-output.txt"

Interactive run (pipe stdin):

printf "6\nATGCAT\n50\n4\n" | ./gradlew run --args="data/motifs-input.txt data/motifs-output.txt"

Other tasks:

./gradlew runDijkstra --args="0 3"
./gradlew runGADijkstra --args="0 3"
./gradlew runKMeans
printf "6\n50\n4\n" | ./gradlew runClustering

Clustering non-interactive:

./gradlew runClustering -Dcluster.motif.length=6 -Dcluster.mutation.percent=50 -Dcluster.motifs.per.sequence=4 \
  --args="data/motifs-input.txt"

Project Layout

• src/main/java/com/motifdiscovery/MotifGAImproved.java: Production entry point
• src/main/java/com/motifdiscovery/MotifGAClassic.java: Reference/legacy algorithm variant
• src/main/java/com/motifdiscovery/GAUtils.java: Minimal GA helpers and input/output utilities
• src/main/java/com/motifdiscovery/clustering/: Optional clustering workflow
• src/test/java/com/motifdiscovery/: Tests
• data/motifs-input.txt: Canonical input file
• data/motifs-output.txt: Canonical output file
• legacy/: Archived experiments, MATLAB, and C implementations

Input Format

The canonical input file is data/motifs-input.txt in a FASTA-like format:

GeneName GeneCode SEQUENCE...

Output

MotifGAImproved writes the predicted motif and consensus to data/motifs-output.txt.

Understanding the Output

• Random candidate motifs: random substrings sampled per sequence; the best fitness is retained.
• Weight matrix: per-position base frequency across the population.
• Pm: predicted motif from highest-frequency bases.
• cons: consensus using IUPAC ambiguity codes.
• POOL: selected parent pool for crossover.
• COUNT: GA iterations until convergence or max iterations.

Test Dataset (Correctness Check)

Use data/test-motifs-input.txt which embeds the known motif ACGTAC in each sequence.

Example run:

./gradlew run --args="data/test-motifs-input.txt data/motifs-output.txt"

Expected behavior:

• Predicted motif should be close to ACGTAC (heuristic, not guaranteed).

Tests include a dataset validation check:

./gradlew test

Notes

• legacy/ is excluded from the Gradle build and is non-production.
• MATLAB code is kept under legacy/matlab for testing only.