A Coupling-Based Periodic Table of Disease Tractability

We introduce a method that assigns every disease a single number — the halving depth D — that predicts whether it responds to single-target therapy or requires drug combinations. The method measures total correlation (TC) in protein interaction networks, decomposes it hierarchically, and determines how many levels of decomposition are needed to contain the coupling. A theoretical capacity line at D = 3.03 separates modular diseases (D < 3.03), where one well-chosen target can control the network, from globally entangled diseases (D > 3.03), where combinations are necessary. We analyze 31 diseases using high-confidence STRING protein interaction data (combined score ≥ 900). Of these, 9 fall below the capacity line and 22 above. All 9 modular diseases have at least one drug class providing clinically meaningful disease control as a single agent. Among global diseases, the most treatment-resistant (cancers, HIV, tuberculosis, autoimmune conditions) uniformly require combinations, while a few with effective replacement or suppressive monotherapies (e.g., insulin for type 1 diabetes) represent a distinct therapeutic mode that the coupling structure does not distinguish from network-level intervention. The method also ranks every protein in each network by its coupling contribution (the "bill"), recovering known drug targets at up to 4.8× enrichment without prior knowledge of drug status. Cross-disease analysis identifies NF-κB (NFKB1) as the most undrugged coupling bottleneck in medicine, ranking in the top 30% across 16 of 31 diseases with no approved inhibitor. All results, code, and data are publicly available at https://github.com/MuffinsThaCat/drug-discovery.