Consumer genetics databases hold dense genotypes of millions of people, and the number is growing quickly. In 2018, law enforcement agencies began using such databases to identify anonymous DNA via long-range familial searches. We show that this technique is far more powerful if combined with a genealogical database of the type collected by online ancestry services. We present a "genealogical triangulation" algorithm and study its effectiveness on simulated datasets. We show that for over 50% of targets, their anonymous DNA can be identified (matched to the correct individual or same-sex sibling) when the genetic database includes just 1% of the population. We also show the effectiveness of "snowball identification" in which a successful identification adds to the genetic genealogical database, increasing the identification accuracy for future instances. We discuss our technique’s potential to enhance law enforcement capabilities as well as its privacy risks.

https://www.biorxiv.org/content/10.1101/531269v1

This project requires both Rust and Python. Populations are generated in Python, then exported to a json file. Rust reads the population file, then runs simulations to generate IBD data, outputting the data to a file. The output is then read in on the python side and fit to generate a model. Identification tasks are performed in Python using the generated model.

Identification will require 16+GB of RAM for populations with generations of 100k individuals.

• Python
• Rust
• GNU scientific library (GSL)

• Python 3.4+
• scipy/numpy
• Cython
• Statsmodels
• progressbar2

• run fetch.sh in data/recombination_rates directory. This will fetch the appropriate recombination data from the HapMap project.
• Install the python dependencies listed above.
• run python3 setup.py build_ext --inplace in the "python" directory to build the cython modules. This will need to be run whenever any of the .pyx files are modified.

Some of the script names aren't descriptive of what is happening, as code changed over the course of the project.

The typical work flow is a three step process

1. Generate population - In the python directory - python3 generate_populatio n.py --help
2. Export population to Rust - In the python directory - python3 export_population.py --help
3. Run simulations in rust - In the rust directory - cargo run --release --bin simulate -- --help
4. Import output from simulations - In the python directory - python3 import_simulation.py --help
5. Identify - In the python directory - python3 evaluate_deanonymize.py --help

To generate a population use the generate_population.py script. For example if you cd into the python directory and run: python3 generate_population.py ../data/your_tree_file ../data/recombination_rates/ --generation_size 1000 --num_generations 10 --output population.pickle A population with 10 generations each with 1000 members will be generated and saved to population.pickle with Python's pickle format. In the paper the generation size was 100,000.

To run experiments in rust first convert the population to a format the simulation can understand.: python3 export_population.py population.pickle file_for_simulation.json --num-anchor-nodes 150

This command will pick 150 nodes from the last three generations and mark them as anchors. It will then output a json file for the rust simulator to read. Then run cargo run --release --bin simulate file_for_simulation.json recombination_directory output_file Then it will perform 1000 experiments to sample from the simulated empirical IBD distributions for the (labeled, unlabeled) pairs. A file output_file will be created with the simulation data. Simulation can run for days, depending on your population parameters and hardware. The simulations from the paper took 20-60 hours to run.

To create the model (ie fit the hurdle-gamma parameters), run import_simulation.py population.pickle work_file --output-pickle model.pickle. The model will be saved to model.pickle.

The final step is identifying individuals.

Running python3 evaluate_deanonymize.py population.pickle model.pickle -n 10 will try to identify 10 random unlabeled individuals in the population.