This repository holds the CrohnsPred model for Genetic Risk prediction of Crohn's Disease.
Trained on 10000+ individual synthetic genomic samples, this model can approximately predict a higher genetic susceptibility to CD.
The prediction tool is integrated into a publicly accessible web-based dashboard, which allows the users to get a personalized prediction and access their score through eye-catching graphics and visualizations for a facile understanding of the results.
Crohn’s Disease (CD) is a chronic illness that has seen a dramatic increase in global prevalence and incidence in the last two decades [3]. The development of Crohn’s has been partially attributed, amongst other factors, to certain mutations in one’s genes [1]. Most current techniques in Bioinformatics focus on investigating how relevant single genetic mutations are in the development of CD and can fall short in considering important interactions between sparse mutations, which are believed to greatly influence the development of complex illnesses like Crohn’s [2].
This project proposes a Deep Neural Network, CrohnsPred, for the prediction of an individual’s Genetic Risk of Crohn’s Disease based on sample-level genotype mutation data. This approach leverages the ability of DNNs to identify patterns and discover links amongst great amounts of data, which have been shown to outperform various statistical models currently used for the polygenic score calculation for other polygenic diseases [5]. To preserve data anonymity and reduce ethical concerns, the genetic and phenotypic data used for the Network training and testing was created as synthetic data through the HAPNEST software [4].
The project is written in python. The following third party packages are required to ensure full project functionality:
- numpy: python package for mathematical and scientific computing.
- pandas: python package for data analysis, particularly suitable for handling relational and labelled data.
- pytorch: python package for machine learning and deep learning algorithms.
All necessary imports, libraries and modules are specified at the top of each .ipynb file.
All genetic data preprocessing is described in the data_preprocessing.ipynb file.
All mutations data from the PGS Catalog is described in the mutations_preprocessing.ipynb file.
The formatting process for the 100k dataset is described in the one_million_dataset_formatting.ipynb file.
To enable prediction, the input .vcf data must include:
- Chromosome (column 'chrom'): an Integer
- Position (column 'pos'): an Integer
- SNP ID (column 'rsid' or 'ID'): as a string 'rsXXYY...'
- Reference Allele (column 'ref'): a character (A/C/T/G) or string
- Alternative Allele (column 'alt'): a character (A/C/T/G) or string
The model architecture, training and testing is available at CrohnsPred.ipynb.
To run the train and test pipelines from scratch, simply acces the up stated .ipynb file and run the cells in order.
Model Hyperparameters
- 'PReLU': hidden layer activation function.
- 'Sigmoid': output activation function.
- 'Adam': optimizer with learning rate = 0.001.
- '20': number of max hidden layers.
- '300': number of epochs.
- '250': training batch size.
The model output will contain a value in the range [0,1], wether a higher risk has not been detected or otherwise.
To use the Web Dashboard created to visualize the model predictions on real data, visit the belfioreasia/WebApp repository.
Belfiore Asia
ec21414
CID:210471618
BSc Computer Science and Mathematics,
Queen Mary University of London
[1] Aleksejs Sazonovs, Stevens, C., Guhan Ram Venkataraman, Yuan, K., Avila, B.E., Abreu, M.T., Ahmad, T., Matthieu Allez, Atzmon, G., Baras, A., Jc, B., Nir Barzilai, Laurent Beaugerie, Beecham, A., Bernstein, Ç.N., Bitton, A., Bernd Bokemeyer, Chan, A., Chung, D.C. and Cleynen, I. (2021). Sequencing of over 100,000 individuals identifies multiple genes and rare variants associated with Crohn's disease susceptibility. medRxiv (Cold Spring Harbor Laboratory). doi:https://doi.org/10.1101/2021.06.15.21258641.
[2] Behravan, H., Hartikainen, J.M., Tengström, M., Pylkäs, K., Winqvist, R., Kosma, V., Mannermaa, A., 2018. Machine learning identifies interacting genetic variants contributing to breast cancer risk: A case study in Finnish cases and controls. Sci. Rep. 8 , 13149. https://doi.org/10.1038/s41598-018-31573-5.
[3] Roda, G., Chien Ng, S., Kotze, P.G., Argollo, M., Panaccione, R., Spinelli, A., Kaser, A., Peyrin-Biroulet, L. and Danese, S. (2020). Crohn’s disease. Nature Reviews Disease Primers, [online] 6(1), pp.1–19. doi:https://doi.org/10.1038/s41572-020-0156-2.
[4] Wharrie, S., Yang, Z., Raj, V., Monti, R., Gupta, R., Wang, Y., Martin, A., O’Connor, L.J., Kaski, S., Pekka Marttinen, Pier Francesco Palamara, Lippert, C. and Ganna, A. (2022). HAPNEST: efficient, large-scale generation and evaluation of synthetic datase ts for genotypes and phenotypes. bioRxiv (Cold Spring Harbor Laboratory). doi:https://doi.org/10.1101/2022.12.22.521552.
[5] Zhou, X., Chen, Y., Ip, F.C.F., Jiang, Y., Cao, H., Lv, G., Zhong, H., Chen, J., Ye, T., Chen, Y., Zhang, Y., Ma, S., Lo, R.M.N., Tong, E.P.S., Mok, V.C.T., Kwok, T.C.Y., Guo, Q., Mok, K.Y., Shoai, M. and Hardy, J. (2023). Deep learning-based polygenic risk analysis for Alzheimer’s disease prediction. Communications Medicine, [online] 3(1), pp.1–20. doi:https://doi.org/10.1038/s43856-023- 00269-x.