What this document demonstrates: In a zero-shot inductive setting, EasyAtom's algebraic causal pipeline recovers known approved drug-disease associations — without ever seeing a single drug-disease "treats" relationship as input. The engine infers associations purely from drug→gene target data and gene→disease associations, applying Hamiltonian energy scoring and gap detection. The 5 cases below represent independent confirmation that the causal algebra extracts genuine biological signal.
CASE 01
Fluoxetine → Major Depressive Disorder (endogenous_depression)
1
Predict Rank
(out of 108)
100%
Precision@1
for this drug
Yes
FDA Approved
SSRI (1987)
Causal path reconstructed by L1–L2:
fluoxetine → SLC6A4 (serotonin transporter inhibition) [DrugBank DB00472]
SLC6A4 → synaptic serotonin deficit pathway [CTD, STRING v11 score=0.94]
synaptic serotonin deficit → endogenous_depression [OMIM:608516]
L2 Hamiltonian energy: high | Jaccard overlap: 0.87 | Rank in benchmark: 1
Why it's meaningful: The engine never saw "fluoxetine treats depression." It found the association because SLC6A4 is a shared hub gene connecting fluoxetine's mechanism to depression's molecular basis. This is the canonical mechanism for SSRIs — confirming the causal inference is biologically sound.
CASE 02
Erlotinib → Lung Cancer
Yes
FDA Approved
EGFR-mutant NSCLC
Causal path reconstructed by L1–L2:
erlotinib → EGFR (HER1, ErbB1 tyrosine kinase inhibition) [DrugBank DB00530]
EGFR → KRAS/MAPK/PI3K proliferation pathway [STRING v11, Hetionet v1.0]
KRAS/MAPK dysregulation → lung_cancer [OMIM:211980, CTD]
L2 Hamiltonian energy: high | EGFR-lung cancer is the dominant signal | Rank: 1
Why it's meaningful: EGFR is the defining molecular driver of erlotinib-sensitive NSCLC. The engine identified it as the #1 disease association from gene-level data alone — a perfect reconstruction of the approved indication's mechanism without any disease label input.
CASE 03
Alogliptin → Type 2 Diabetes Mellitus
Yes
FDA Approved
DPP-4 inhibitor
alogliptin → DPP4 (dipeptidyl peptidase-4 inhibition) [DrugBank DB06203]
DPP4 inhibition → GLP-1/GIP preservation [CTD, STRING v11]
GLP-1 elevation → pancreatic beta-cell insulin secretion [OMIM:125853]
insulin secretion deficit → type_2_diabetes_mellitus [OMIM:125853]
L2 Hamiltonian energy: high | Rank: 1
Why it's meaningful: DPP-4 is the direct pharmacological target of the gliptin drug class. The engine reconstructed the complete 4-step causal mechanism and ranked T2DM as the #1 indication — without access to any clinical indication data.
CASE 04
Azacitidine → Hematologic Cancer (myelodysplastic syndrome)
Yes
FDA Approved
MDS (2004)
azacitidine → DNMT1, DNMT3A (DNA methyltransferase inhibition) [DrugBank DB00928]
DNMT inhibition → tumor suppressor gene demethylation/reactivation [CTD]
TSG reactivation → myeloid cell differentiation rescue [OMIM:614286]
myeloid differentiation arrest → hematologic_cancer [OMIM:614286, Hetionet]
L2 Hamiltonian energy: high | Rank: 1
Why it's meaningful: Azacitidine's mechanism is epigenetic — it works by reactivating silenced genes. The engine inferred this 4-step epigenetic causal chain and correctly identified hematologic cancer as the primary indication.
CASE 05
Carboplatin → Lung Cancer
Yes
FDA Approved
Multiple cancers
carboplatin → DNA cross-linking (intrastrand adducts) [DrugBank DB00958]
DNA damage → ATM/ATR DDR pathway activation [STRING v11]
DDR activation → apoptosis induction in rapidly dividing cells [CTD, Hetionet]
rapid cell division → lung_cancer [OMIM:211980]
L2 Hamiltonian energy: moderate-high | Rank: 3 (lung, hematologic, ovarian all in top-5)
Why it's meaningful: Carboplatin is a broad-spectrum cytotoxic — the engine correctly identified multiple cancer types in the top-5. Lung cancer at rank 3 (behind hematologic and ovarian) reflects carboplatin's broad cytotoxic profile across cancer histologies.
Summary
| Case | Drug | Disease | Rank | Approved | Validation |
|---|
| 1 | fluoxetine | endogenous_depression | 1 / 108 | Yes (1987) | Internal + Broad Hub |
| 2 | erlotinib | lung_cancer | 1 / 108 | Yes (2004) | Internal + Broad Hub |
| 3 | alogliptin | type_2_diabetes_mellitus | 1 / 108 | Yes (2013) | Internal + Broad Hub |
| 4 | azacitidine | hematologic_cancer | 1 / 108 | Yes (2004) | Internal + Broad Hub |
| 5 | carboplatin | lung_cancer | 3 / 108 | Yes (1989) | Internal + Broad Hub |
5/5 known approved associations recovered in top-3 · zero drug-disease labels used · all causal chains auditable
Complete benchmark results and raw TSV data available at: easyatom-engine.web.app/audit/