Skip to content

About

Motif and motif-interaction discovery from DNA/RNA/protein sequence-to-expression data via game-theoretic attribution of a trained neural network.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

97 Commits

Folders and files

Repository files navigation

MotifInference.jl

Dev

Discover functional motif combinations from sequence-to-measurement data (DNA / RNA).

📖 Documentation: kchu25.github.io/MotifInference.jl — including the full run_method argument reference.

Installation

using Pkg
Pkg.add("MotifInference")

A CUDA-capable GPU is required for training.

Usage

Put your data in a CSV with two columns (sequence, expression), one row per sequence:

sequence,expression
GATCACAGGTCTATCACCCTATTAACCACTCACGGGAGCTCTCCATGCATT,1.5
TTGACAGCTAGCTCAGTCCTAGGTATTATGCTAGCTACTAGAGAAAGAGGA,2.0
CGTACGATCGATCGTAGCTAGCTAGCATCGATCGATCGTACGTAGCATCGA,-0.5
AATTGGCCAATTCCGGAATTCCGGTTAACCGGTTAACCGGAATTCCGGAAT,0.8
TGCATGCATGCATGCATGCATGCATGCATGCATGCATGCATGCATGCATGC,1.2

Then point run_method at it:

using MotifInference

run_method("/path/to/mydata.csv")

That's the whole pipeline: load, tune, train, attribute, render.

Where results are saved

Results (trained model, motif caches, and HTML reports) are written to a folder named after the CSV, created under save_root (the current directory by default):

<save_root>/<csv basename>/

So run_method("data/mydata.csv") writes to ./mydata/. Control the location with two keyword arguments:

# choose the parent directory
run_method("data/mydata.csv"; save_root="/path/to/results")   # -> /path/to/results/mydata/

# choose the folder name explicitly
run_method("data/mydata.csv"; save_folder_name="run01")        # -> ./run01/

Options

# RNA sequences, fixed seed (skips hyperparameter tuning)
run_method("/path/to/mydata.csv"; seq_type=:rna, seed=42)
Option Default Meaning
seq_type :dna :dna or :rna
seed nothing RNG seed; nothing triggers hyperparameter tuning
save_root "." where the results folder is created

A header row is auto-detected and skipped; blank lines are ignored. Sequences of differing length are padded to the max length before encoding.

About

Motif and motif-interaction discovery from DNA/RNA/protein sequence-to-expression data via game-theoretic attribution of a trained neural network.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages