Model description

The locus-to-gene (L2G) model derives features to prioritise likely causal genes at each GWAS locus based on genetic and functional genomics features. The main categories of predictive features are:

    - Distance: (from credible set variants to gene)
    - Molecular QTL Colocalization
    - Variant Pathogenicity: (from VEP)

    More information at: https://opentargets.github.io/gentropy/python_api/methods/l2g/_l2g/

Intended uses & limitations

[More Information Needed]

Training Procedure

Gradient Boosting Classifier

Hyperparameters

Click to expand

Hyperparameter	Value
objective	binary:logistic
base_score
booster
callbacks
colsample_bylevel
colsample_bynode
colsample_bytree	0.8
device
early_stopping_rounds
enable_categorical	False
eval_metric	aucpr
feature_types
feature_weights
gamma
grow_policy
importance_type
interaction_constraints
learning_rate
max_bin
max_cat_threshold
max_cat_to_onehot
max_delta_step
max_depth	5
max_leaves
min_child_weight	10
missing	nan
monotone_constraints
multi_strategy
n_estimators
n_jobs
num_parallel_tree
random_state	42
reg_alpha	1
reg_lambda	1.0
sampling_method
scale_pos_weight	0.8
subsample	0.8
tree_method
validate_parameters
verbosity
eta	0.05

How to Get Started with the Model

To use the model, you can load it using the LocusToGeneModel.load_from_hub method. This will return a LocusToGeneModel object that can be used to make predictions on a feature matrix. The model can then be used to make predictions using the predict method.

    More information can be found at: https://opentargets.github.io/gentropy/python_api/methods/l2g/model/

Citation

https://doi.org/10.1038/s41588-021-00945-5

License

MIT