Configuration

For the train and infer commands, sstar2 requires a YAML configuration file specifying parameters for simulation, preprocessing, and the machine learning model (GradientBoostingRegressor). An example configuration file can be found here and is shown below:

simulation:
  nrep: 10000
  nref: 5
  ntgt: 10
  ref_id: YRI
  tgt_id: CEU
  seq_len: 50_000
  mut_rate: 1.29e-8
  rec_rate: 1e-8
  ploidy: 2
  is_phased: false
  nfeature: 100_000
  is_shuffled: false
  nprocess: 2
  seed: 12345

preprocessing:
  vcf_file: examples/data/sstar2.example.biallelic.snps.vcf.gz
  chr_name: "21"
  ref_ind_file: examples/data/ref.samples.list
  tgt_ind_file: examples/data/tgt.samples.list
  win_len: 50_000
  win_step: 10000
  is_phased: false
  nprocess: 2
  ploidy: 2

model:
  params:
    loss: "quantile"
    alpha: 0.99
    n_estimators: 200
    max_depth: 3
    random_state: 12345 

The configuration file has three top-level sections: simulation, preprocessing, and model.

In the simulation section:

Parameter	Description
nrep	Number of simulations run in each batch. Additional batches are simulated until at least nfeature genomic windows are obtained.
nref	Number of reference individuals.
ntgt	Number of target individuals.
ref_id	Population label used for the reference population in the simulated data.
tgt_id	Population label used for the target population in the simulated data.
seq_len	Length of the simulated sequence.
mut_rate	Mutation rate used in simulation.
rec_rate	Recombination rate used in simulation.
ploidy	Ploidy of the simulated individuals.
is_phased	Whether the simulated data are treated as phased.
nfeature	Minimum number of genomic windows to generate for model training.
is_shuffled	Whether the generated genomic windows are shuffled before model training.
nprocess	Number of processes used for simulation.
seed	Random seed for reproducibility.

In the preprocessing section:

Parameter	Description
vcf_file	Path to the input VCF file for inference.
chr_name	Chromosome name to analyze.
ref_ind_file	Path to the file listing reference individuals.
tgt_ind_file	Path to the file listing target individuals.
win_len	Window length used to divide the genome.
win_step	Step size between adjacent windows.
is_phased	Whether the input VCF is treated as phased.
nprocess	Number of processes used for preprocessing.
ploidy	Ploidy of the individuals in the input VCF.

In the model section, parameters under params are passed to GradientBoostingRegressor:

Parameter	Description
loss	Loss function used by the gradient boosting model. Must be set to quantile.
alpha	Quantile level used.
n_estimators	Number of boosting stages.
max_depth	Maximum depth of each regression tree.
random_state	Random seed used by the model.

For other available model parameters, please see the official scikit-learn documentation for GradientBoostingRegressor.