Codebase for the Computational Functional Genomics course offered in January 2026 at IISER Pune.

The goal of the project is to build a classifier to identify regions in the genome that are binding sites for transcription factors(TFs).

The first part of the project involves building a Markov-based classifier and testing it via k-fold cross validation. We use Precision-Recall and Receiver-Operator metrics.

Dependencies

Python (3.13.7)

• NumPy (2.2.6)
• PanDas (3.0.0)
• MatPlotlib (3.10.3)
• tqdm (4.67.3)
• pyfaidx (0.9.0.3)

Data

Since whole genome data is too huge to upload here, you'll have to download the FASTA files. Here's how to do that:

For MacOS

• curl -o projectData/chr1.fa.gz https://hgdownload.soe.ucsc.edu/goldenPath/hg38/chromosomes/chr1.fa.gz
• gunzip projectData/chr1.fa.gz (replace chr1 with whichever chromosome from 1-19)

For Linux

• wget -P projectData/ https://hgdownload.soe.ucsc.edu/goldenPath/hg38/chromosomes/chr1.fa.gz
• gunzip projectData/chr1.fa.gz

For Windows

• Install curl and do the same as above.

Running the code

The code can be executed via the terminal using the command:
python run_kfold.py --tf_id [TF_ID] --chr_id [CHR NUMBER] --markov_order [M] --kfold [K] --num_cpus [#CORES USED]

• chr_id has to be of the form chr<n> where n is a number in [1,19].
• tf_id has to one of [REST, EP300, CTCF].
• kfold has to be a number in [1,10].
• markov_order has to a number in [0,10].

To run the simpler version of the code:
python3 simplerVersion.py --markov_order [M] --fasta_path [RELATIVE PATH]

.csv files containing data on sensitivity, specificity, precision will be stored in /resultsData.
/projectData contains FASTA files for the chromosomes as well as bound-unbound information for each chromosome segrated into bins.

Precision-Recall Curves will be stored in ./PRC_PLOTS and Receiever-Operator Curves in ./ROC_PLOTS.