Neur-Ally

Neur-Ally is a deep learning model for predicting the regulatory effect of neurologic single nucleotide variations

Dependencies

Python version = 3.8.16
OS = Ubuntu 20.04.4

Python Libraries

tensorflow (2.7.0) (Tensorflow dependencies recommended for the specific version is required for GPU support)
numpy (1.23.1)
matplotlib (3.4.2)
seaborn (0.11.1)
pandas (1.3.1)
pyfaidx (0.6.2)
adjustText (0.8)
liftover (1.1.16)
scikit-learn (1.2.2)

Required Files

Human reference genome (hg38.fa)
Encode Exclusion files ("GRCh38_unified_blacklist.bed","dukeExcludeRegions.bed")
Encode epigenomic processed narrowpeak bed files for training input data generation
GWAS Catalog dataset (gwas_catalog_all_v1.0)
dbSNP common variant dataset (common_all_20180418.vcf)
GTEx portal v8 signif_variant_gene_pairs.txt files.
"gencode_exons_modif_canonical.bed" (bed file from UCSC with canonical exon coordinates according to hg38 build)
"ALL.wgs.phase3_shapeit2_mvncall_integrated_v5c.20130502.sites.vcf" (1000 genome phase3 vcf file)
"encode_ccre.bed" (bed file from UCSC with encode candidate cis-regulatory coordinates according to hg38 build)
"FASTA_BED.ALL_GRCh37.novel_CUPs.bed" (bed file with conversion-unstable position coordinates, can be downloaded from "https://github.com/cathaloruaidh/genomeBuildConversion".)

Scripts

Neur-Ally can be trained on epigenomic datasets and used for variant effect prediction.

Scripts for preprocessing, training and testing

The scripts for running the model are:

dataprocess.py

--Creates training, validation and testing dataset from encode .bed files

neurally_train.py

--Trains the model on training data.

neurally_test.py

--Testing the model after training. The weights files for the 27th epoch are provided in the repository which can be directly used for testing the model.

Scripts for variant effect prediction

The scripts for predicting the regulatory effect of non-coding variants:

gwasall.py and gwaspos.py

--gwasall.py outputs all GWAS Catalog variant coordinates to file, gwaspos.py outputs neurologic GWAS variant coordinates to file as the positive variant set

eqtlall.py and eqtlpos.py

--eqtlall.py outputs all eQTL variant coordinates to file, eqtlpos.py outputs top 1000 neurologic eQTL variant coordinates to file as the positive variant set

negset.py and negfile.py

-- negset.py outputs negative non-functional SNPs to file and negfile.py saves sad-score array of negative non-functional SNP set to .npy file

asdpos.py

--outputs Autism Spectrum Disroder(ASD) GWAS variant coordinates to file

sadscores.py

--Variant effect prediction using in-silico mutagenesis, creates output file with SAD scores.

evalue.py and signifeval.py

--evalue.py outputs e-value of positive SNPs and signifeval.py writes significant e-values of positive SNPs to file

Scripts for plotting the results

heatmap_eval.py and manhattan_eval.py

--heatmap_eval.py plots heatmap of SNPs with significant e-values and manhattan_eval.py plots manhattan plot of SNPs with significant e-values

Usage

Running the model

git clone https://github.com/anilprakash94/neur_ally.git neur_ally

cd neur_ally

Then, run the programs according to the requirements and instructions listed in README.md.

For example:

python3 dataprocess.py -h

usage: dataprocess.py [-h] [-d DATASET_PATH] [-e FILE_LIST]
                      [--excl_files EXCL_FILES] [-i INPUT_BINS]

Neur-Ally Data Preprocessing

optional arguments:
  -h, --help            show this help message and exit
  -d DATASET_PATH, --dataset_path DATASET_PATH
                        path of folder containing encode datasets
  -e FILE_LIST, --file_list FILE_LIST
                        specify the file to which encode filenames are written
  --excl_files EXCL_FILES
                        path of bed files having exclusion regions
  -i INPUT_BINS, --input_bins INPUT_BINS
                        specify the file name to which preprocessed input data
                        is written