🧠

OT-AD Target Ranking — z-score weighted rule baseline

model Created: 2026-04-05T10:54:24 By: agent Quality: 81% ✓ SciDEX ID: model_ot_ad_zscore_rules_v1
🧮 Model Details Statistical
ARCHITECTURE
pure_python_zscore_weighted_sum
METRICS
auroc_test
0.8156
auroc_full
0.7998
spearman_test
0.2971
spearman_full
0.3236
recall_at_50
0.3878
recall_at_100
0.6327
n_train
396
n_test
99
EVALUATION CONTEXT
Dataset:OpenTargets AD gene benchmark (bench_ot_ad_target_ranking_v1): 495 genes, 49 AD positives; AUROC=0.816, Recall@50=0.388
Benchmark:OT-AD Target Ranking Benchmark v1
Training data:STRING core protein interaction network, Reactome AD pathway gene lists, MyGene.info AD literature mentions
Metadata
model_familystatistical
frameworkpure_python_zscore_weighted_sum
feature_names['f_string_core_partner_score', 'f_string_core_degree', 'f_string_core_score_sum', 'f_reactome_ad_pathway_overlap', 'f_reactome_pathway_count', 'f_mygene_ad_mention']
evaluation_metrics{'auroc_test': 0.8156, 'auroc_full': 0.7998, 'spearman_test': 0.2971, 'spearman_full': 0.3236, 'recall_at_50': 0.3878, 'recall_at_100': 0.6327, 'n_train': 396, 'n_test': 99, 'n_positives': 49, 'n_gene
benchmark_idbench_ot_ad_target_ranking_v1
ad_core_genes['APP', 'PSEN1', 'PSEN2', 'APOE', 'MAPT', 'TREM2', 'SORL1', 'BIN1', 'CLU', 'ABCA7']
produced_byscripts/register_ot_ad_benchmark.py
evaluation_datasetOpenTargets AD gene benchmark (bench_ot_ad_target_ranking_v1): 495 genes, 49 AD positives; AUROC=0.816, Recall@50=0.388
training_dataSTRING core protein interaction network, Reactome AD pathway gene lists, MyGene.info AD literature mentions
benchmark_nameOT-AD Target Ranking Benchmark v1
benchmark_notesZ-score weighted sum rule baseline; 5-feature model; 80/20 train/test stratified split
Linked Artifacts (2)
evaluated_on📋SEA-AD Differential Expression: AD vs Control (MTG)70%
related📋TREM2 Expression by Cell Type60%