Code and workflows developed by the overall winning team at the 2nd International BioHackathon organized by the Bioinformatics Italian Society (BITS) (Padova, Italy — May 26–27, 2026).
- Overall Winning Team — 2nd International BioHackathon (BITS 2026)
- Honorable Mention — Reverse Engineering of Gene Regulatory Networks challenge
Team Members:
- Rida Hayat
- Alessandro Leone
- Margherita Morocutti
- Dhyani Solanki
Official award document: WinningTeam.pdf
The hackathon focused on reverse engineering biological networks from high-throughput omics data across two independent tracks, each evaluated on 10 benchmark datasets with ground truth followed by blind validation datasets.
Biological Data ──┬──► Task 1: Metagenomics ──► Compositional transforms (CLR/TSS) ──► Bacterial Co-occurrence Network
│
└──► Task 2: Transcriptomics ──► Random Forest regression (GENIE3) ──► Gene Regulatory Network (GRN)
- Challenge: Metagenomic taxonomic count tables are high-dimensional, zero-inflated, and strictly compositional (constrained simplex). Direct Pearson correlation on raw counts introduces severe spurious associations.
- Workflow:
- Filtering & Preprocessing: Low-abundance filtering and minimal positive offset zero-imputation.
- Compositional Transformations: Total Sum Scaling (TSS) and Centered Log-Ratio (CLR) transformations to break compositional dependencies.
- Network Inference: Ridge-regularized precision matrix estimation (graphical Gaussian modeling) and rank correlation across taxa.
- Evaluation: Threshold scanning against gold-standard adjacency matrices using AUROC and precision-recall metrics.
- Challenge: Reconstructing directed transcriptional regulatory networks from gene expression matrices with simulated knockouts, knockdowns, and multifactorial perturbations.
- Workflow:
- Model: Implemented GENIE3 (GEne Network Inference with Ensemble of trees). For each target gene, a Random Forest regressor predicts its expression profile using all remaining genes as candidate regulators.
- Confidence Scores: Variable importance measures from the ensemble trees serve as directed edge weights.
- Threshold Optimization: Scanned thresholds across 10 benchmark datasets, identifying 0.035 as the optimal cutoff balancing precision, recall, and F1 score.
- Blind Prediction: Applied the calibrated model to generate binary adjacency matrices for the 3 blind validation datasets.
BiohackathonBITS2026/
├── README.md # Project documentation and summary
├── WinningTeam.pdf # Official winning team certificate
├── BioHackathon_2026_Instruction.pdf # Competition instructions and task rules
├── Transcriptomics.Rmd # Task 2 solution (GENIE3 GRN pipeline)
├── Metagenome.Rmd # Task 1 solution (Compositional network pipeline)
├── Trascriptomics_Dataset/ # Task 2 expression & gold-standard matrices
│ ├── Dataset1/ ... Dataset10/
│ └── validation/
└── MetaGenomics_Dataset/ # Task 1 taxonomic counts & adjacency matrices
├── dataset_1/ ... dataset_10/
└── validation/
Both pipelines are self-contained R Markdown notebooks.
Make sure R (>= 4.0) is installed. Required libraries:
# Bioconductor
if (!requireNamespace("BiocManager", quietly = TRUE)) install.packages("BiocManager")
BiocManager::install("GENIE3")
# CRAN packages
install.packages(c("data.table", "ggplot2", "pheatmap", "PRROC", "reshape2", "dplyr"))Open and knit the notebooks in RStudio, or render from terminal:
# Task 1: Metagenomics pipeline
Rscript -e "rmarkdown::render('Metagenome.Rmd')"
# Task 2: Transcriptomics (GRN) pipeline
Rscript -e "rmarkdown::render('Transcriptomics.Rmd')"- Organizers: Giacomo Baruzzo, Giulia Cesaro, Matteo Pallocca (Bioinformatics Italian Society - BITS)
- Event: BITS 2026 Hackathon