mutation_prediction¶

Purpose¶

Runs model-based mutation classification from merged features and produces VCF outputs.

Upstream¶

• merge_feature

Required inputs¶

• combine_feature_parquet (all_feature.parquet from merge_feature)

Input interpretation¶

Input key	Source step	Required	Interpretation
combine_feature_parquet	merge_feature output	Yes	Integrated feature matrix used for model inference; schema must match model expectations.
model_dir	config (steps.mutation_prediction)	Yes	Directory containing trained model artifacts for prediction.
model_name	config (steps.mutation_prediction)	Yes	Selected model artifact name/version under model_dir.

Parameters¶

Effective runtime keys (current implementation)¶

Parameter	Location	Type	Interpretation
model_dir	steps.mutation_prediction	path string	Directory containing trained model artifacts used for inference.
model_name	steps.mutation_prediction	string	Model identifier/name loaded from model_dir.

Common pretrained model names shipped in SpaceTracer_new_github/models:

• spatial_free_model
• spatial_feature_preserved_model

Present in template but not fully wired in current step runner¶

Parameter	Location	Note
random_seed	steps.mutation_prediction	Template key exists; current step uses internal constant seed.
plot	steps.mutation_prediction	Template key exists; current step uses internal plotting setting.

Tuning notes¶

• Ensure model artifacts match the feature schema in combine_feature.
• Keep model versioning explicit (model_dir + model_name) for reproducibility.
• If extending this step, expose internal hardcoded options as config keys progressively.

Outputs¶

• raw_pred_vcf: <step_dir>/results/Sample_total_pred_truesites.vcf
• final_vcf: <step_dir>/results/Sample_total_pred_truesites_PASS.vcf