BITHub is an web resource that allows exploration of gene expression across eight curated large-scale transcriptomic datasets of the human post-mortem brain. It integrates, harmonises, and standardises data from multiple studies to enable consistent cross-dataset exploration of gene-level expression patterns across brain regions, developmental stages, and clinical contexts.
BITHub is designed to support both interactive biological exploration and reproducible computational analysis.
bithub/
├── data-preprocessing/ # Dataset-specific cleaning and harmonisation
├── pipeline/ # Data packing, scaling, and transformation
├── frontend/ # Interactive web interface
└── README.md
data-preprocessing/: Dataset-specific scripts to clean, harmonise, and annotate raw expression and metadata files.
pipeline/:Unified processing pipeline that transforms preprocessed data into analysis-ready formats used by the frontend (e.g. z-score normalisation, gene filtering, dataset packing).
frontend/ The interactive web application that powers BITHub, allowing users to search and explore expression patterns across datasets.
BITHub currently integrates eight curated, large-scale transcriptomic datasets derived from human post-mortem brain tissue. These datasets span multiple brain regions, developmental stages, and disease contexts, and were selected based on cohort size, data quality, and metadata availability.
The expression files were pre-processed using the code in the data-preprocessing folder, whereas the data packing pipeline, including the z-score transformations are in the pipelines folder.
This stage includes:
- Cleaning and formatting raw expression matrices
- Curating and filtering metadata
- Defining anatomical regions and developmental stages
- Removing low-quality or non-informative samples and features
This stage converts preprocessed datasets into a common format used by the frontend:
- Gene-level filtering and alignment across datasets
- Z-score transformation within datasets
- Packaging of expression and metadata into lightweight, queryable objects
This separation ensures that raw data handling decisions are clearly distinguished from downstream transformations.
Step 1: Clone the repository
https://github.com/VoineaguLabUNSW/bithub.git
Step 2: Open data pre-processiong
cd bithub/data-preprocessing
Step 3: Change the config file
All dataset-specific input and output paths for preprocessing are defined in data-preprocessing/config/paths-example.yaml.
Before running any preprocessing scripts update the entries so they match the locations of your local/raw data files and desired output directories.
Step 4: Run the preprocessing for the invidual datasets
All preprocessing and exploratory analyses are implemented as R Markdown notebooks located in
data-preprocessing/notebooks/.
These notebooks must be run in the following order, as each step depends on outputs generated in the previous one:
-
metadata-preprocess.Rmd
Cleans, filters, and harmonises metadata across datasets, including variable selection and annotation standardisation. -
bulk-deconvolution.Rmd
Performs bulk tissue deconvolution using the harmonised metadata and expression matrices to estimate cell-type composition. -
drivers-of-variation.Rmd
Identifies major technical and biological sources of variation in the data and evaluates their impact on gene expression patterns.
In addition, two R Markdown files are provided for figure generation:
- A notebook for main figures
- A notebook for supplementary figures
Each of these has a corresponding pre-rendered .html file for easy inspection without rerunning the analysis.
All notebooks read input paths from the configuration file defined in
config/paths-example.yaml.
Step 5: Run the data packing pipleine Once the expression and metadata matrices are generated, run the data packing pipeline [In progress - for Kieran to complete]
BITHub supports flexible gene-centric exploration, including:
Selection of datasets for comparison Stratification by brain region Stratification by developmental stage Filtering based on available clinical or technical metadata These options allow users to explore both dataset-specific patterns and cross-study consistency.